Reduced vowel production in American English among Spanish ...

Florida International UniversityFIU Digital Commons

FIU Electronic Theses and Dissertations University Graduate School

12-3-2012

Reduced vowel production in American Englishamong Spanish-English bilingualsEmily ByersFlorida International University, [email protected]

DOI: 10.25148/etd.FI12120519Follow this and additional works at: https://digitalcommons.fiu.edu/etd

This work is brought to you for free and open access by the University Graduate School at FIU Digital Commons. It has been accepted for inclusion inFIU Electronic Theses and Dissertations by an authorized administrator of FIU Digital Commons. For more information, please contact [email protected].

Recommended CitationByers, Emily, "Reduced vowel production in American English among Spanish-English bilinguals" (2012). FIU Electronic Theses andDissertations. 800.https://digitalcommons.fiu.edu/etd/800

FLORIDA INTERNATIONAL UNIVERSITY

Miami, Florida

REDUCED VOWEL PRODUCTION IN AMERICAN ENGLISH AMONG

SPANISH-ENGLISH BILINGUALS

A thesis submitted in partial fulfillment of the

requirements for the degree of

MASTER OF ARTS

in

LINGUISTICS

by

Emily Byers

2012

ii

To: Dean Kenneth G. Furton College of Arts and Sciences This thesis, written by Emily Byers, and entitled Reduced Vowel Production in American English among Spanish-English Bilinguals, having been approved in respect to style and intellectual content, is referred to you for judgment. We have read this thesis and recommend that it be approved.

Tometro Hopkins Jean-Robert Cadely Mehmet Yavas, Major Professor

Date of Defense: November 2, 2012 The thesis of Emily Byers is approved.

Dean Kenneth G. Furton College of Arts and Sciences Dean Lakshmi N. Reddi University Graduate School

Florida International University, 2012

iii

© COPYRIGHT 2012 by Emily Byers

All rights reserved

iv

ACKNOWLEDGMENTS

I would like to express my heartfelt gratitude to Dr. Mehmet Yavas for his

attention and guidance throughout the duration of this project. His continued support and

encouragement has been vital to the successful completion of this thesis.

I would also like to extend sincere thanks to the other two members of my thesis

committee, Dr. Tometro Hopkins and Dr. Jean-Robert Cadely, for their valuable advice

and commitment of time to this project. I have been extremely fortunate to have the

benefit of not only an excellent committee, but also the council of very supportive

members of the linguistics faculty.

Special thanks go to Dr. Paulette Johnson for her assistance in processing the

results of this study. I also wish to extend a personal thanks to Dr. James Sutton for his

generous support of my research and to the English Department for providing an

academic environment of the highest quality.

v

ABSTRACT OF THE THESIS

REDUCED VOWEL PRODUCTION IN AMERICAN ENGLISH AMONG SPANISH-

ENGLISH BILINGUALS

by

Emily Byers

Florida International University, 2012

Miami, Florida

Professor Mehmet Yavas, Major Professor

Prominent views in second language acquisition suggest that the age of L2

learning is inversely correlated with native-like pronunciation (Scovel, 1988; Birdsong,

1999). The relationship has been defined in terms of the Critical Period Hypothesis,

whereby various aspects of neural cognition simultaneously occur near the onset of

puberty, thus inhibiting L2 phonological acquisition. The current study tests this claim of

a chronological decline in pronunciation aptitude through the examination of a key trait

of American English – reduced vowels, or “schwas.” Groups of monolingual, early

bilingual, and late bilingual participants were directly compared across a variety of

environments phonologically conditioned for vowel reduction. Results indicate that late

bilinguals have greater degrees of difficulty in producing schwas, as expected. Results

further suggest that the degree of differentiation between schwa is larger than previously

identified and that these subtle differences may likely be a contributive factor to the

perception of a foreign accent in bilingual speakers.

vi

TABLE OF CONTENTS

CHAPTER PAGE

CHAPTER 1. Introduction and Background ...................................................................... 1 1.1. Vowel Reduction in American English ........................................................... 1

1.1.1. Prototypical phonetic qualities of schwa .......................................... 3 1.1.2. The [ɨ ] / [ə] distinction in SAE vowel reduction............................. 5 1.1.3. Vowel quality distinctions between plural and possessive

morphemes ...................................................................................... 7 1.1.4. “Targetlessness” in word-internal schwa production ....................... 8 1.1.5. Other properties of schwa – duration and prosody ......................... 11 1.1.6 An introduction to prominent voices in the field of language

acquisition ..................................................................................... 12 1.2 CPH: The “Critical Period Hypothesis” ......................................................... 13 1.3 Flege’s “Speech Learning Model” .................................................................. 19 1.4 Kuhl’s “Native Language Magnet Model” ..................................................... 23 1.5 Contemporary Issues in Reduced Vowel Analysis ......................................... 25

CHAPTER 2. Research Design and Methodology ........................................................... 27 2.1 Participants ...................................................................................................... 29 2.2. Materials - Instrument I ................................................................................. 31

2.2.1. Plural Morpheme ............................................................................ 31 2.2.2. Possessive Morpheme ..................................................................... 31 2.2.3. Reduced Vowels in Word-Final Position ....................................... 32

2.3. Instrument II................................................................................................... 33 2.3.1. “Deletable Schwa” Instrument ........................................................ 34 2.3.2. “Non-Deletable Schwa” Instrument ............................................... 35

2.4 Procedure ........................................................................................................ 36 2.5 Explanation of the procedure .......................................................................... 37 2.6 Methods of Analysis ....................................................................................... 38

CHAPTER 3. Results and Discussion .............................................................................. 41 3.1 Experiment 1 ................................................................................................... 42

3.1.1. Experiment 1(a) Vowel Qualities Comparison ............................... 42 3.1.2. Experiment 1(b) Reduced vowel duration: A cross-group

comparison .................................................................................... 49 3.1.3 Effects of preceding consonants on vowel duration in word-

final position ................................................................................. 53 3.1.4. Effects of segmental features on the vowel qualities of

reduced vowels in word-final position .......................................... 57 3.2. Experiment II ................................................................................................. 67

3.2.1. Comparison of reduced vowel lengths in word-internal environments ................................................................................. 69

3.2.2. Semantically-related words ............................................................. 72 3.3. Summation of the results ............................................................................... 73

vii

CHAPTER 4. Conclusion ................................................................................................. 74 4.1. Theoretical Implications for Reduced Vowel Production ............................. 74

4.1.1. Speech Learning Model .................................................................. 74 4.1.2. Native Language Magnet Model .................................................... 76

4.2 Generalizations regarding L2 reduced vowel production ............................... 77 4.2.1. Vowel quality distinctions .............................................................. 77 4.2.2. Variation in schwa duration ............................................................ 79 4.2.3. Semantically-related word pairs ..................................................... 80

4.3. Limitations and Extensions ............................................................................ 82

References ......................................................................................................................... 84

APPENDICES .................................................................................................................. 87

viii

LIST OF FIGURES

FIGURE PAGE Figure 1.2.a Timeline of the relationship between AOA and linguistic competency ....14

Figure 3.1.a Across-groups comparison of F1 values in plural, possessive and word-final environments ............................................................................45

Figure 3.1.b Mean duration for schwa in CV# position in monolinguals ......................54

Figure 3.1.c Mean duration for schwa in CV# position in early bilinguals ...................55

Figure 3.1.d Mean duration for schwa in CV# position in late bilinguals .....................56

Figure 3.1.e Across-groups comparison of F1 values to post-labial CV# environments ..............................................................................................58

Figure 3.1.f Across-groups comparison of F1 values to post-coronal CV# environments ..............................................................................................59

Figure 3.1.g Across-groups comparison of F1 values of post-dorsal CV# environments ..............................................................................................60

Figure 3.1.h Descriptive statistics of F2-F1 positioning across groups ........................62

Figure 3.1.i Comparison of the position of schwa in CV# position among monolinguals ..............................................................................................64

Figure 3.1.j Comparison of the position of schwa in CV# position among early bilinguals ....................................................................................................65

Figure 3.1.k Comparison of the position of schwa in CV# position among early bilinguals ....................................................................................................66

Figure 3.1.l Across-groups comparison of the average values among three places of articulation ..................................................................................67

Figure 3.2.a Across-groups comparison of duration in deletable and non-deletable schwas.........................................................................................70

Figure 4.2.a Prototypical F1 and F2 values for SAE vowels .........................................78

ix

LIST OF TABLES

TABLE PAGE

Table 3.1.a Pairwise comparison of F1 values across monolinguals ...........................42

Table 3.1.b Pairwise comparison of F1 values across early bilinguals .........................43

Table 3.1.c Pairwise comparison of F1 values across late bilinguals ...........................44

Table 3.1.d Mean values of F2-F1 for monolinguals, early bilinguals, and late bilinguals ....................................................................................................48

Table 3.1.e Pairwise comparison of average durations in monolingual speakers ........49

Table 3.1.f Pairwise comparison of average durations in early bilinguals ...................51

Table 3.1.g Pairwise comparison of average durations in late bilinguals .....................52

Table 3.1.h Across-categories comparison of word-final vowels in late bilinguals ....................................................................................................57

Table 3.2.a Comparison of deletable and nondeletable vowel durations across populations .................................................................................................71

Table 3.2.b Paired sample T-tests of semantically-related word pairs .........................73

x

LIST OF SYMBOLS

SYMBOL DESCRIPTION

/ / Indicates Phoneme

[ ] Indicates Phonetic Representation

C Consonant

V Vowel

# Indicates Word Boundary

ə/ʌ Mid Central Lax Unrounded Vowels

ɨ High Central Lax Unrounded Vowel

ɛ Mid Front Lax Unrounded Vowel

aɪ Diphthong Moving from Low Back to High Front Vowel

ɔ Mid Back Lax Rounded Vowel

ɪ High Front Lax Unrounded Vowel

ŋ Velar Nasal Consonant

ð Voiced Interdental Fricative Consonant

ɻ Rhotic Retroflex “r” Consonant

æ Low Front Lax Unrounded Vowel

ɑ Low Back Tense Unrounded Vowel

Voiceless Dental Stop Consonant

1

C H A PT E R 1. I ntr oduction and B ackgr ound

1.1. V owel R eduction in A mer ican E nglish

Schwa production is a frequently occurring phonological process in American

English which results from the neutralization of multiple vowel quality contrasts

(Chomsky and Halle, 1968). The process of vowel reduction in stress-timed languages is

rule-governed in that it is restricted to unstressed syllables, which may then undergo a

secondary phonological process of deletion in selected environments. Fokes (1993)

targets the rhythm of a language, particularly the intervals at which stressed beats fall, as

a primary factor in determining which vowels are candidates for reduction. Schwas are

distinct from what are typically classified as “fast speech” reductions (see Dalby, 1986)

where full vowels experience a centralization in both their height and backness features

(and can also be optionally deleted) as a result of a rapid speech rhythm, subsequently

rendering reduced vowels in both content and function words as in the following

examples:

“I’m going to the store” [aɪ m gɔ ɪ ŋ tə ðə stɔɹ ]

“What do you mean?” [wɑ də ju min]

In typical conversation, as in the aforementioned examples, a speaker strives to produce

utterances which are sufficiently intelligible to the hearer while maximizing economy in

the production (Harrington, 2010). These schwa productions are more likely to occur

when the discourse level is informal, particularly when a speaker feels it is unnecessary

to pay special attention to clarity of speech. A further contributive factor to the

2

production of schwas in fast speech may be the statistical frequency with which the target

word is used in the language. Wright (2003) theorizes that the vowels of “hard” words

are pronounced in an expanded vowel space as compared with those of “easy” words.

The “hard”/ “easy” distinction is correlated with a language’s neighborhood density,

whereby words possessing a high value neighborhood density exist alongside many other

words containing obvious phonemic similarities. “Easy” words, therefore, meet the

requirements of occurring frequently in the language while possessing a low

neighborhood density – thus decreasing the likelihood that the target word will be

confused with another (see Harrington, 2010). Fast-speech schwa productions, however,

should not be classified alongside the “definite”1

Through systematic analysis of the acoustic properties of schwa production in

monolinguals, the current study addresses the question: How is this mid-central lax vowel

produced in the oral cavity? As chapters two and three will demonstrate, measurements

of the features “vowel height” and “backness” of schwas across a variety of

phonologically-conditioned environments provide a framework of average schwa values

for specific phonological settings. It is necessary from the onset to introduce

comprehensive background information regarding generalizations that can be made about

the prototypical phonetic characteristics of schwa in order to provide a basic

representation of schwa in the American

vowel system, which is the focus of the current study, as definite schwa production elicits

the reduction of a full vowel at any speech tempo (i.e., “Carolina” [kɛɹ ə laɪ nə],

“christening” [kɹ ɪ sənɪ ŋ]).

11 This term is often used in the literature to indicate a phonological process as

contrasted with hypoarticulation of a vowel.

3

understanding of the perceptual and articulatory challenges that Spanish-English

bilinguals face when presented with target schwa, as compared with the Spanish vowel

system.

1.1.1. Pr ototypical phonetic qualities of schwa

As is to be expected, properties of schwa (such as duration and vowel qualities)

show high variation among different demographics. Factors such as age, gender, and

regional dialect all contribute to markedly different sound productions. Other factors

which affect schwa production include the method in which constructions are elicited

(i.e., reading a sentence vs. oral interviews) and even the level of intimacy between

speaker and listener. Though these social factors certainly create high levels of acoustic

variation in reduced vowel production, both between speakers and within the same

speaker across utterances, it is not impossible to make legitimate generalizations

regarding the phonetic characteristics of schwa.

In a monolingual Standard American English speaker (SAE hereafter), a schwa is

the result of the movement of a vowel away from the periphery into the underutilized

central vowel space. Chomsky and Halle (1968) consider these contrasts to be

“neutralized” because the effect results in the features [- high, - low] and [- front, - back].

Table 1.1.a provides an illustration of the position of schwa in the vowel inventory of

Standard American English.

4

Table 1.1.a Vowels of Standard American English

Sour ce: www.mnsu.edu/phonetics/SA E

Schwa is unique among the vowels of SAE because, unlike /ʌ /, it is not phonemic

in nature. That is, while /ʌ / is able to contrast minimally with other vowels (/bæt/ vs.

/bʌ t/), schwa does not. The unconscious nature of the vowel reduction leaves speakers

without a mental representation for [ə], as their L1 inventory leads the untrained ear to

believe that it is actually peripheral vowels which are being produced. The contrast in

perception and production becomes increasingly problematic for L2 learners whose L1

vowel inventories contain neither central vowels, nor reduction processes, as the result is

often peripheral vowel substitution in lieu of the target phoneme. Table 1.1.b illustrates

the Spanish vowel inventory, including the allophone [ɛ ] for /e/.

5

Table 1.1.b Vowels of Spanish

Sour ce: T he Sounds of Spanish: A nalysis and A pplication by R ober t H ammond, 1999.

As demonstrated in Table 1.1.b above, the Spanish vowel inventory contains no

phonemes within the mid-central or high-central regions. Therefore, L2 target vowels

which fall into this otherwise barren region of the oral cavity will be considered reduced

in that they fall into a bilingual’s “reduced vowel zone.” Table 1.1.a is a particularly

detailed chart of the SAE vowel system in that it includes both phonemic and allophonic

representations which are not always listed in other accounts of the SAE vowel inventory.

Of particular interest is the [ɨ ] / [ə] distinction, about which there have been studies (see

Flemming, 1997, 1999: Hammond 1999) both advocating for, and dismissing, the need

for separate transcription symbols for these reduced central vowel sounds.

1.1.2. T he [ɨ ] / [ə] distinction in SA E vowel r eduction

An issue which is well-established in the literature concerning vowel reduction is

the predictive nature of F1 and F2 formants (indicating vowel height and the degree of

backness) in environments phonologically conditioned for schwa. Hammond (1999) and

Flemming (2009), among others, identify a distinction in the phonetic properties of [ə]

and [ɨ ], though their conclusions differ in what should be done regarding transcription

and analysis.

6

For Hammond, the distinction is of an articulatory nature. He describes “barred-

i” [ɨ ] as a high, central, lax, unrounded vowel which is clearly articulated in a higher

manner than the mid, central, lax, unrounded vowel transcribed as schwa [ə]; however,

Hammond argues for the general acceptability of a “reduced vowel zone.” In essence, he

argues that since there is nothing contrastive about the [ɨ ] / [ə] distinction, emphasis

should be placed upon whether or not the L2 speaker is able to move a peripheral vowel

into the otherwise unoccupied central vowel space. Furthermore, it is the case that

monolingual speakers may alternate between [ɨ ] and [ə] depending on whether they are

speaking a regional dialect or SAE, thus causing both sounds to be judged as acceptable

by native speakers. Therefore, it is the shared opinion of the current study that [ɨ ] and

[ə] should both be judged as accurately reduced vowels, occupying the mid-to-high

central “reduced vowel zone,” with no need to further dissect the zone into central and

high-central areas when making acceptability judgments in the analysis. It follows then

that the transcription of all reduced vowels as [ə] is acceptable for the binary distinction

“reduced/nonreduced” as presented in chapters two and three.

Flemming, however, calls for a distinction in the phonetic transcriptions of mid-

central lax [ə] and high-central lax [ɨ ] in order to better match the phonetic properties of

these centralized vowels in context. His call for two distinct transcriptional markers is a

consequence the fact that reduced vowels in the plural morpheme are impressionistically

and instrumentally different from schwas in both the possessive morpheme and word-

finally. It has been observed that these differences in the vowel qualities of “schwa”

occur predictably when certain morphemes are compared against other phonological

environments. Flemming and Johnson (2007) found that schwas in word-final position

7

(i.e., “china” [ʧaɪ nə], “comma” [kɑmə]) have a relatively consistent mid-central vowel

quality. In monolingual speakers, the values of F1 formants (indicating vowel height) in

word-final schwas rank consistently higher than those in bound morphemes, indicating

that word-final schwa is produced with a lower vowel. In Flemming’s (2009)

monolingual participants, the average F1 value for word-final schwa was 665 Hz, which

is considerably higher than the 500 Hz that Olive et al. (1993) identified as a standard F1

value for schwa in Standard American English.

According to Flemming et al., the vowel qualities of the plural morpheme {-əz}

(as in “places”) differ significantly from two other environments containing a schwa

nucleus: word-final schwa [ə] and the possessive morpheme {-əz} (as in “Chris’s”),

which Flemming concludes is simply the construction of ‘word-final schwa + /s/’.

Though the plural and possessive morphemes are transcribed, and are written in the

orthography, identically, Flemming and Johnson (2007) prove that they are not

homophonous with regard to vowel quality production. The minimal pair “Rosa’s roses,”

originally noted by Trager and Bloch in 1941 (Flemming and Johnson, 2007), illustrates

the failings of general IPA transcriptions to account for the obvious differences in the

pronunciations of the aforementioned words. There is certainly an argument to be made

that transcribing these sounds as separate symbols would elicit more native-like

pronunciations of words than the blanket symbol [ə] is able to represent.

1.1.3. V owel quality distinctions between plur al and possessive mor phemes

Flemming and Johnson identify the plural morpheme schwa (as in “roses”) as

containing a lower F1, thus creating a vowel higher in the vocal tract, than those in

possessive morphemes (such as “Rosa’s.) Their study concludes that typical values one

8

expects to find for the vowel qualities “height” and “backness” in the possessive

morpheme {-əz}should be expected to mirror those of schwa in word-final endings.

Essentially, Flemming and Johnson argue that schwa in the possessive morpheme is in

actuality word-final schwa with only the addition of possessive marker {-s} (realized

phonetically as [-z]) to differentiate the environment. The testing of this claim, among

monolinguals and two bilingual populations, is one of the central focuses of my study.

Flemming and Johnson further distinguish schwas in the plural morpheme {-əz} from

possessive and word-final environments by arguing that reduced vowels in the plural

morpheme are more suitably grouped with other word-internal schwas (as in

[prɪ nsɨ pəl]), which they claim are better represented by [ɨ ]. The claim becomes

problematic in view of Flemming’s (2009) work concerning the “targetlessness” of word-

internal schwa.

1.1.4. “ T ar getlessness” in wor d-inter nal schwa pr oduction

The transparent ability of word-internal schwa to assimilate its vowel qualities to

neighboring consonants has led researchers (Kondo 1994; Flemming 2009) to pose the

question: Is medial schwa targetless with regard to F1 and F2 values? Observations that

schwa is articulated closer to the front of the oral cavity in anticipation of a labial sound,

or higher in anticipation of an alveolar or palate-alveolar, have fostered the notion that

medial reduced vowels may serve as empty place holders, devoid of independent vowel

qualities, in word-internal environments. Since 1960’s linguists have determined that

surrounding consonant environments temper vowel qualities by shifting vowel formants

toward more central values (Stevens and House, 1963). These studies report the largest

9

effects on F2 values, the measure which indicates the relative front/back position of a

vowel in the oral cavity.

More recently, Browman and Goldstein (1990), Kondo (1994) and Flemming

(2009) have all conducted empirical studies regarding the question of whether or not

reduced vowels exhibit patterns under spectrographic analysis. Specifically, they seek to

determine whether schwa’s F1 and F2 formants have any predictable target values.

Kondo, in particular, cites numerous arguments in favor of the approach that medial

schwas, rather than striving for the (F1)500/ (F2)1500/ (F3) 2500 Hz ,which are numeric

representations of ideal centralization, are actually at the mercy of assimilatory processes

caused by segments surrounding the nucleus. According to Kondo’s analysis, both vowel

centralization and contextual assimilation (caused by adjacent consonants) imply target

undershoot2

The results of Kondo’s (1994) experiment indicate that average schwa values for

F1 formants did indicate a target vowel height, and that only F2 values (relative front vs.

back positioning) could be considered “targetless.” These results implies that, while

vowels are susceptible to moving forward or backward in the mouth in anticipation of

upcoming consonants, the height values remain relatively fixed regardless of the

. Kondo suggests two possible results stemming from this undershooting:

either the vowel will not reach its target and will succumb to the characteristics of

surrounding segments or the value will miss its target and instead will find a more

centralized value.

2 For further information on vowel undershoot see Lindblom’s (1963) mathematical

model of vowel reduction, which identifies vowel undershoot as a function of vowel duration.

10

upcoming segment. The debate is far from closed, however, as studies by Hillenbrand et

al. (2001) found that consonant environment had only a small, though statistically

significant, effect on vowel intelligibility. Browman and Goldstein (1990) provide a

supplemental explanation for the inherent variability of medial schwas, whereby certain

phonological segments, specifically schwas, may overlap in time (or be “co-produced) in

order to reflect the combined articulatory influences of adjacent segments.

A second hypothesis, which is compatible with Kondo’s findings, is that a given

phonetic unit may be unspecified for a particular dimension. If the target phone is

unspecified for a dimension, most likely its F2 value, then it is expected to find that the

resultant value under spectrographic analysis will be a result of the movement from the

preceding segment, across the valueless nucleus, to the value of the following segment.

From a phonological standpoint, Browman and Goldstein are confirming that it is

possible to have medial schwa be completely unspecified for tongue position, whereas

Kondo suggests that only the feature [+back] may be unspecified.

Flemming (2009) identifies the average F1 (height) values of medial schwa at 428

Hz, which is significantly lower than the values for word-final schwa. Such a

discrepancy leads Flemming to support the aforementioned ideas concerning schwa’s

tendency to assimilate to the surrounding context. Flemming and Johnson (2007) prove

this intuition empirically through the use of spectrographic analysis to determine, within

their instrument, that higher F2 vowels were found in environments where schwa is

adjacent to coronal consonants. In essence, this study found that schwas were produced

closer to the front of the mouth when the following consonant contains a [+coronal]

feature.

11

A final point of interest with regard to the phonetic properties of reduced vowels

is raised by Flemming (2009) when he suggests “…a lesser degree of assimilation to

context [in addition to the hypothesis that medial schwas undergo assimilation of height,

backness, and lip position] still results in raising of low vowels because of assimilation to

the narrow constrictions of adjacent consonants” (18). Here Flemming suggests that

medial reduced vowels, by virtue of their location inside the root, are physically unable to

escape constriction of the narrowed vocal tract should they be encased between

obstruents.

1.1.5. Other pr oper ties of schwa – dur ation and pr osody

In addition to prototypical vowel qualities, there are certain other generalizations

about schwa production that, given a sufficient corpora, can be extracted from linguistic

studies. Another characteristic which reliably differentiates schwa from peripheral

vowels is duration – schwas are much shorter than their full vowel counterparts.

Flemming (2009) identifies medial schwa duration at an average of 64 milliseconds for

monolingual SAE speakers, compared with 153 milliseconds if the schwa occurs in

word-final position. Hammond (1999) identifies the following two part pattern of SAE

vowels: 1) Stressed vowels are much longer than unstressed vowels and 2) Unstressed

vowels are always realized as reduced vowels. This claim will be examined in the course

of the current study.

In addition to generalizations about duration, there are several parameters which

researchers have set regarding the vowel qualities of schwa. Olive et al. (1993) identify

schwa, produced in a neutral articulatory position, as having formant frequencies which

occur at equal intervals positioned at roughly 500 Hz (F1), 1500 Hz (F2), and 2500 Hz

12

(F3). In reality, schwa is not often produced in a neutral articulatory position, which

results in less than uniform, evenly-spaced formants. This is the result of suprasegmental

effects of English. Schwa is not produced in a vacuum, but rather is the result of vowel

reduction processes brought forth by patterns of unstressing pretonic and posttonic vowel

nuclei, which is predictable in SAE. Even though “ideal” schwas with evenly spaced

formants rarely surface in medial position, an identification of the prototypical (ideal)

schwa allows researchers to describe L2 speakers’ attempts at vowel quality

neutralization and identifies areas in which bilingual populations show variation from

native speakers – a key contribution to the perception of a foreign accent.

1.1.6 A n intr oduction to pr ominent voices in the field of language acquisition

Flemming and Johnson (2007) have previously identified patterns in vowel

quality production which are predictable based on the particular schwa-containing

morpheme. True schwas provide testable variables (F1 and F2 values) which allow the

current study to make practical application of two theoretical models of language

acquisition: Flege’s Speech Learning Model (SLM) and Kuhl’s Native Language

Magnet model (NLM). Both models attempt to account for universal trends in language

acquisition apart from a biological reliance on the critical period hypothesis, though

neither model denies the existence of a relationship between age of acquisition (AOA) of

L1/ L2 and the accuracy of a speaker’s phonological output.

Before Flege’s SLM and Kuhl’s NLM can be discussed in the context of AOA, it

is necessary to provide adequate background information regarding the “critical period

hypothesis” (CPH hereafter). Section 1.2 describes CPH in detail, as well as noting the

evaluations several prominent experts in the field of L1 and L2 acquisition have made

13

regarding the connection between neurobiological processes and phonological

inventories. Having established the principle theory (CPH) which Flege and Kuhl refine

through their respective speech models, sections 1.3 and 1.4 examine the central tenets of

the SLM and NLM, noting that these two theories are functionally compatible in that

NLM is primarily concerned with infants’ L1 acquisition, whereas SLM works

extensively to explain variable success rates in L2 acquisition, not dismissing age as a

crucial predictor of success. Section 1.5 incorporates additional prominent observations

(Valdman, 1976; Strange, 2008) regarding the role of perception and production in L2

acquisition – without the inclusion of these voices a thorough review of the literature

concerning phonological acquisition would remain incomplete.

1.2 C PH : T he “ C r itical Per iod H ypothesis”

The critical period hypothesis (CPH) has its roots in the observation by Lenneberg

(1969) that as humans mature neurologically, the capacity to acquire language is

diminished in some way. Even in long-term acquisition of an L2 phonological system,

that is, child learners will be expected to exceed adult learners in the mastery of new

sounds. Patkowski (1994) formalizes the CPH as follows:

There is a period, ending around the time of puberty (operationally defined

to mean somewhere between the ages of 12 and 15 years), during which it

is possible, but not inevitable, for learners to acquire, as an end-product of

a naturalistic L2 acquisition process, full native-like fluency in the

phonological system of a second language, and after which such a

possibility does not exist anymore… (206)

14

A central tenet of the CPH is the notion that the critical period is related to an unspecified

neurological change which essentially causes a loss of elasticity in the brain as linguistic

sound systems become more “fixed” with regard to language acquisition. This process

has also been described by many researchers as a loss of plasticity in the brain which co-

occurs with the completion of hemispheric lateralization in the brain (Ioup, 2008).

It is important to reemphasize here that the inability of adults to acquire native-

like L2, as explained by the CPH, refers solely to the pronunciation of native-like

phonemes. That is, it is entirely possible to continue developing other outstanding

abilities in L2 acquisition such as sophisticated syntax or the ability to acquire a rich and

varied lexicon. For the purposes of this study, the term “linguistic output” will be used to

refer only to the production of sounds and prosodic features. Figure 1.2.a provides an

illustration of the CPH’s postulation of the relationship between AOA and L2 linguistic

output.

Figure 1.2.a Timeline of the relationship between AoA and linguistic competency

Source: T. Scovel (1988) A Time to Speak: A Psycholinguistic Inquiry into Critical Period for Human Speech from M. Yavas’ (2011) Applied English Phonology.

15

Lenneburg’s observation was originally intended to explain both L1 acquisition and cases

involving aphasia; however, noticeable disparities concerning phonetic accuracy in child

versus adult L2 learners led researchers to question whether AOA is the determining

factor in achieving native-like pronunciation. Best and Tyler (2007) articulate the issue

succinctly as follows:

Prior contact with the stimulus language, and position along the trajectory

of native or first language (L1) development, converge in some crucial

way to shape one’s perception of phonetic details and phonological

structure in speech. (14)

Traditionally, the “critical period” (or “sensitive period” in some works), after which

insurmountable obstacles are faced in L2 phonological acquisition, has been set at the

“onset of puberty,” typically taken to mean twelve years of age (see Flege, 2005).

Alterations to this critical age include Long’s very specific addendum to the CPH that an

L2 is spoken without a foreign accent if learned by the age of six, with a foreign accent if

learning begins after age twelve, and with variable success in between (from Birdsong,

1999). Long equates these stages to the maturational stages infants and children progress

through in traditional L1 acquisition, which exist independently of cognitive development

(see Patkowski, 1994). An insistence that language acquisition is apart from cognitive

development may explain why adults may surpass the comparative rate at which children

acquire phonology in the very earliest stages of L2 phonological acquisition, and why

adults are predicted to outperform children throughout the course of L2 learning with

16

regard to the non-phonological aspects of a language (i.e. syntax, lexical acquisition,

etc.).

It is true that many people who acquire a second language in adulthood speak

with an easily detectable foreign accent. Flege and Hillenbrand (1984) attribute this

difference in part to phonetic distances between the speaker’s native language and target

language, and also to the speaker’s ability to manipulate the phonological inventories of

the L1 and L2 systems. In these cases, post-critical or “late” bilinguals resort to

producing L2 words with at least some of their L1 phonemic inventory in a process

known as “equivalence classification,” thus creating a complicated interlanguage

phonology which causes the “foreign accent” so easily detectable by the untrained ear

(see Flege, 1987; Strange and Shafer, 2008). Discourse surrounding issues of perception

attempts to explain the mental processes of multilanguage acquisition; however; current

perception models acknowledge that AoA remains the most prominent factor in acquiring

a native-like sound system.

Patkowski (1990, 1994) further advocates for the existence of a critical period;

however, he maintains that in order to reliably and legitimately test the CPH it is

imperative that research designs be longitudinal in nature. Long-term studies are

required so that adult participants are given time to reach the level of “optimal learning”.

Specifically, the post- critical period L2 learner must have had both prolonged exposure

to the L2 and live in an environment where there is daily use of the L2 in order to have

comparable input to that which children receive in their L2 acquisition process (from

Birdsong, 1999).

17

Patkowski (1994) concedes that IQ, or advanced cognitive development, plays an

enthusiastic role in facilitating L2 acquisition during adulthood. Nevertheless, Patkowski

maintains that the role played by the input child L2 learners receive, assuming the

prototypical child acquires language in a primarily immersive environment, undoubtedly

serves children to achieve better rates of fluency and more native-like pronunciation than

should be expected of adult L2 learners. Flege has taken issue with the empirical data on

which Patkowski’s (1990) conclusions rest; however, he acknowledges that the

differences in findings could be the result of conflicting methodological considerations or

different manners of statistical analysis, rather than a full discrediting of Patkowski’s

findings. Disagreements concerning methodological approaches may affect the way L2

acquisition data is collected and presented, yet both researchers agree that there is an age-

related period which serves to enhance the capabilities of L2 phonological learning.

While pinpointing AoA as the most prominent factor affecting the degree of

foreign accent, Flege has maintained that AoA cannot be the only cause of foreign accent,

or a lack thereof, in L2 speakers. Flege et al. (2006) point to the fact that L2 acquisition

prior to the end of the critical period does not equip L1 Korean children with native-like

pronunciation as proof that L2 foreign accent can be affected by many factors apart from

age. Most prominent among these non-biological factors is the amount of L1 use (Flege

et al., 1997), while factors such as L1 proficiency, length of residence, and gender have

yet to be adequately proven or invalidated (see Piske et al., 2001).

Since there is more to language acquisition than a simple “pre-critical

period”/”post-critical period” binary split, Flege’s SLM provides a gradient scale toward

native-like productions that is considerably more flexible than biological explanations

18

allow for. While acknowledging that age is a, if not the, necessary condition for native-

like phonetic competency, Flege argues that the biological processes needed to acquire

new language remain intact throughout the lifespan and that it is actually perception, not

production mechanisms, which cause phonetic errors in adults (Ioup, 2008). Flege’s

argument against a set cutoff for the optimal age of acquisition is that no specific age has

shown a particular drop off in linguistic aptitude when tested empirically.

One shortcoming of the critical period hypothesis is that it is not predictive of all

L2 learners – specifically, the identification of a critical period does not explain why

some post-critical L2 learners fare much better than others at mastering L2 phonemic

systems, nor does it explain why not all children achieve native-like pronunciation if

exposed to an L2 during the pre-critical stage. In Flege and Hillenbrand (1984), the

researchers support the validity of the proposed “critical period” but refute the rigidity of

the hypothesis based on a belief that, given environmental conditions comparable to those

encountered during L1 acquisition (particularly a full-emersion environment in which the

speaker has an urgent need to be understood), the L2 speaker can continuously improve

upon the native-like pronunciation of target phonemes without ever reaching an impetus.

Ioup (2008) identifies this continual L1 input (that is, an environment where the L2 is not

the dominant daily language) as a contributive factor in the failure of “early” bilinguals to

achieve native-like L2 pronunciation.

Linguistic environments such as this may explain the results of Flege’s studies of

Korean-English and Italian-English bilinguals, which led him to the conclusion that the

CPH cannot be the only predictive factor which is sustained empirically in large data

samples because a sizeable number of “pre-critical” bilinguals continue to carry a foreign

19

accent. At any rate, it should be remembered that Flege’s SLM and Kuhl’s NLM model

are attempts to complement, and not invalidate, the CPH.

Most researchers agree that, even in L2 acquisition, there exists a biological

window before adolescence during which time the acquisition of a second language

produces more phonetically accurate target sounds. It is the purpose of the current study

to determine whether early learners, some of whom are English-dominant whereas others

continue to receive large amounts of L1 input at home, are more capable of perceiving

and producing reduced vowels in intersentential environments than those who learned

English after the critical period.

1.3 F lege’ s “ Speech L ear ning M odel”

One of the most commonly referenced theoretical models of cross-linguistic

perception is Flege’s (1995a) Speech Learning Model (SLM hereout). Concerned with

linguistic acquisition and development over the course of L2 acquisition (measured in

years, not months), the SLM navigates the pitfalls of L1 interference and incorrect

perceptual judgments on the part of the L2 speaker. At the heart of the SLM is the

concept that, though not without certain difficulties, the linguistic faculties involved in

language acquisition remain present throughout a person’s lifespan – namely, there is no

physiological reason why adults cannot acquire a language as rapidly, and with as much

flexibility, as children do (see Ioup, 2008).

Flege’s model primarily addresses issues involving phonetic production, and later

on perception, of L2 phonemes. In addition to outlining the framework by which a

speaker’s L1 is perceptually categorized in the brain, the SLM also grapples with issues

of variable success rates in producing native-like L2 phonemes. In his theories regarding

20

perception, Flege acknowledges the difficulties in perceiving new L2 sounds after the

solidification of an L1 phonemic system. These difficulties arise because the incoming

L2 phonemes are not mapped into a vacuum, but rather must be included, and

distinguished, from a fully-formed L1 phonological system which crowds out sounds

whose production difference is more subtle (i.e., “dental t” [ ] versus “alveolar t” [t]).

Flege also concedes that, while adults retain the capacity to map and store new phonemes

just as children do, the process is inhibited by these four factors: cross-linguistic

similarity, age of arrival, frequency of L1 usage, and the storage of L1 and L2 categories

in a community space, as mentioned above (Escudero, 2007).

One point which seems counterintuitive to the casual observer’s perception of

language is that it seems easier for L2 learners to acquire sounds they perceive as

“foreign” over sounds they perceive to be “similar.” Flege postulates that the phonemes

acquired in L1 and L2 share a mutual phonemic inventory space from which target

sounds may be retrieved. Therefore, if a sound is perceived as existing in a previously

occupied section of a speaker’s phonemic inventory, the new target sound is perceived as

a positional allophone and the L1 phonological system subsequently filters out

information which identifies the target phoneme as a unique L2 sound to be acquired in a

process known as “equivalence classification.” (see Yavas, 2011). It follows that the

greater difference between sounds relates directly to perception of the new phoneme due

to the fact that phonetically distinct sounds will be perceived more readily than sounds

which share multiple phonetic properties. Flege et al. (1981, 1984, 1987, 1997, 2005)

write extensively on the effects of new and similar phones as evidence for a system of

equivalence classification.

21

The SLM model maintains that adult learners encounter difficulty processing or

categorizing L2 phonemes if information highlighting the contrastive properties in L1

and L2 sounds is susceptible to filtering processes, or can be a “nonlearned sound” due to

perceived similarity. It follows then that sounds which differ markedly from the L1 with

regard to place and manner features will form new phonological categories in the brain

and be more easily identifiable, and therefore, recoverable.

It is an aim of the SLM is to present a comprehensive explanation of the strategies

L2 speakers employ when presented with phonemes not included in the L1 inventory.

Therefore, the levels of L2 phonemic identification can be trisected into three categories:

“identical,” “similar,” and “new.” Identical sounds pose no problem for the L2 learner,

who simply transfers L1 feature information onto the L2 phoneme. “Similar” sounds

often lead the L2 speaker to retrieve features reserved for L1 phonemes in an effort to

produce sounds which are “close enough” to the target phoneme through a process of

assimilation, while allowing for a minimal expansion of the phonemic inventory. “New”

sounds contribute to the formation of previously nonexistent phonological categories,

meaning the speaker is conscious that a new sound, complete with a unique feature set,

has been acquired which is identified as “foreign” when compared to the L1 phonemic

inventory.

Escudero (2007) defines sound perception, as realized in the SLM, as follows:

“…the discerning of the phonetic features or properties in the signal that make it possible

to identify the appropriate ‘positional-defined allophones’ or ‘phonetic categories’ of the

language” (122). Flege’s aim, in the development of SLM, is to determine whether

certain L2 segments are either unlearnable, given a speaker’s L1 phonetic inventory, or

22

learnable only by children as the CPH would suggest. The tendency of SLM is to refute

any target sound as completely unlearnable, though perceptual difficulties may take years

of exposure to overcome. Bohn and Munro (2007) pinpoint a fundamental postulate of

SLM as the assertion that “the mechanisms and processes used learning the L1 sound

system remain intact over the lifespan” (5).

The distinction between certain features of separate languages, such as VOT in

English versus Spanish, becomes more complicated in early bilinguals. Researchers

believe that, rather than being in possession of two separate phonologies, early bilinguals

often assign a compromise value to similar sounds in an effort to appease the

requirements of both languages (Flege, 1991). The production of these voiceless stops in

early bilinguals is able to satisfy the voicing requirement for intelligibility in English, yet

consistently measures as far shorter duration in aspiration than one finds in monolingual

English speakers. An explanation offered by the SLM is that when an L2 phoneme is not

perceived as being a separate sound from the L1 inventory, presumably because of the

similarity with a preexisting segment, the two segments will merge to create one whose

values approximate the difference in the target sounds.

One final aspect of the SLM which is crucial to understanding the development of

L2 phonologies deals with the acquisition of new phonemes when an introduced sound is

perceived as separate from the L1 phonological system. Flege (1995a) hypothesizes that

when a new segment is acquired in the L2 inventory, some speakers may undergo an

unconscious process of dissimilation, in order to further distance the new sound from

preexisting sounds in the oral cavity in order to maximize phonemic contrast. This

dissimilation process may result in an overdifferentiation between L1 and L2 phonemes,

23

which the SLM believes will be gradually corrected, or refined, throughout the course of

the language learning process. For this reason, later additions to the SLM advocate for

viewing dissimilation between L1 and L2 segments as a continuum rather than the

originally proposed tripartite structure from the 1984-1993 model.

1.4 K uhl’ s “ Native L anguage M agnet M odel”

The Native Language Magnet Model of language acquisition conceives of

phoneme acquisition as a complex set of phonological abstractions, a markedly different

approach from the phonetic compromises and target undershoots previously discussed.

According to Kuhl (2000), infants learn the perceptual mappings of their ambient

language. Her Native Language Magnet Model (NLM hereafter) highlights the

emergence of neural maps in order to perceive speech signals. First the infant is exposed

to all of the L1 phonemes through constant contact with caregivers, which the infant

perceives as “signals.” As these signals are eventually decoded, the representations

become stored in the memory (known as a “mapping process”) where they form the basis

of linguistic production (Escudero, 2007). These mapping processes are specific for each

individual language, an impediment to linguists attempting to uncover universal mapping

systems which could apply to all languages. Kuhl (2000) further proposes that infants’

perception of sounds becomes categorical in nature due to statistical processing, which

results in a patterning of the acoustic parameters of speech sounds somewhere between

the ages of six and twelve months.

The mapping of acoustical parameters into a classification of “sounds” can be

extremely problematic for second language learners who are attempting to map the new

language onto the appropriate phonetic categories. This is a result of the fact that these

24

idealized representations have become anchored in the brain, causing a perceptual

interference when non-native phonemic categories are presented (Ioup, 2008).

Furthermore, the existence of this filter for the purpose of capturing sounds, which

contains a relatively complete “native language” by the age of twelve months, makes

learning an another language as an adult extremely difficult. As Flege suggested,

learning “similar” sounds also becomes an issue because the prototype of a given

phoneme acts as a magnet – thus forcing the hearer to perceive the new phoneme as the

prototype (Ioup, 2008). The Native Language Magnet Model seems to refute certain

tenets of the CPH on the basis that the onset of puberty is not responsible for neurological

changes, but rather it is implementation of initial mental mappings onto the neurological

pathways which create barriers to accepting foreign phonemes later on in life. Escudero

(2007) suggests the major underdevelopment of NLM exists in its failure to provide a

physical account of how the perceptual mappings for two languages are separated in the

brain, nor does the NLM explain how the activation of overlapping regions in the brain is

achieved in young children. The NLM does not account for language acquisition

successes outside of L1 acquisition in infancy, though it does explain how the process of

filtering out non-native sounds begins.

Kuhl’s hypothesis would seem to exclude the fact that children who learn another

language after one year of age still acquire the L2 phonemic system much easier than

adults. Kuhl suggests in response that early in life it is possible to acquire multiple

neurological mappings because interference effects (from the L1 representations) are

considered to be minimal. For adults, she hypothesizes, it is impossible to have

overlapping regions of the brain when processing two languages, though this is not

25

necessarily the case when bilingualism is acquired early in life. This would seem to serve

more as a modification of the CPH than as a purely non-linguistic explanation of

language acquisition.

1.5 C ontempor ar y I ssues in R educed V owel A nalysis

Having established the theoretical frameworks through which the data in the

current study will be analyzed, it is worthwhile to highlight several of the major

instrumental studies which have previously been conducted on bilingual populations.

Numerous longitudinal studies (Flege and Bohn, 1989; Flege, Munro, and Fox, 1994;

Flege, Bohn, and Jang, 1997; Flege and MacKay, 2004, et al.) have been conducted

regarding L2 phonological achievement across different age (and language) groups.

Thompson (1991) collected speech samples of Russian immigrants, both as free speech

and reading samples, and rated the productions for accentedness (see Patkowski, 1994).

Among a wide variety of independent variables Thompson tested, AoA was the most

reliable predictor of a perceivable accent. It is important to mention, however, that

participants who arrived in the United States before the end of the critical period were

judged to have a lesser accent than those who came as adults, which is not synonymous

with the claim that the child learners were devoid of accent.

“Accentedness” tests a monolingual population’s reactions to L2 production in

non-native speakers. However, testing the role of perception in the L2 speaker must be

conducted through a separate course of study. One of the most common methods used to

test L2 speaker perception is by testing consonants or vowels which do not contrast in the

speaker’s L1 inventory. Flege and MacKay (2004) perform exactly this experiment when

testing the ability of Italian L2 English learners’ abilities to discriminate between /ɒ / -

26

/ʌ /, /ɛ / - /æ/, and /i/ - /ɪ /. From this study, Flege and MacKay were able to confirm that

two conditions were necessary for native-like vowel discrimination: the participant must

have acquired English before the onset of the critical period and also must not be

operating in an L1-dominant environment. That is, participants who categorized

themselves as high-use Italian speakers scored significantly lower on the discrimination

test than participants who, for all intents and purposes, are L2 dominant.

In determining perceptual abilities across Spanish-English bilinguals, Flege,

Munro, and Fox (1994) found that abilities to differentiate between English vowels is not

so much the product of length of exposure to the English vowel inventory as the ability to

perceive dissimilarity is dependent upon the amount of space which lies between vowels

in question. These findings contain interesting implications for the current study on

vowel reduction because it poses the question: will Spanish-English bilinguals be more

capable of perceiving reduced vowels in instances where the full vowel (which

underwent reduction) extends further to the periphery? Will L2 speakers be able to hear

the difference in their interphonological full vowel substitutions, or will perception biases

filter out the acoustic properties of SAE’s schwa substitutions? Another study of interest

to the present group of participants is Flege and Bohn’s (1989) investigation of L1

Spanish speakers’ pronunciation of morphophonological alternations in English words

(i.e. “able” vs. “ability”) – however, the size of the instrument in this study would have to

be greatly increased in order to access generalizations about Spanish-English bilinguals

as a population.

The role of L2 perception in making discrimination judgments in English vowels

has been thoroughly established across a wide variety of L1 groups. The current seeks to

27

contribute to the ongoing study of L2 phonological production by examining vowel

reduction in Spanish-English bilinguals across eight distinct phonological environments.

These environments have been identified as reliable environments to test measurable

characteristics of target reduced vowels. Analysis of F1 and F2 values in the target

schwas, as well as duration, will uncover patterns regarding the production of reduced

vowels in SAE in both early and late Spanish-English bilinguals, the results of which will

contribute to a better understanding of the perception and production errors L2 SAE

speakers make when faced with a language which employs an unusually large vowel

inventory and unfamiliar prosodic features that must be overcome if the speaker is to

develop the stress-timed syllable structure which characterizes American English.

28

CHAPTER 2. Research Design and Methodology

The previous chapter identified a significant gap in the research pertaining to

reduced vowel productions in L2 SAE speakers. Being home to a large and diverse

population of bilinguals who operate in both languages during their daily lives, Miami

lends an opportunity to research schwa production in people whose L1 (or one of their

L1s) does not allow for reduced vowel production. The present study draws from

bilinguals who, because of prolonged exposure to SAE, are capable of producing schwas,

as well as participants who have varying levels of success in producing reduced vowels.

This second group, termed “late bilinguals,” may experience success reducing vowels in

certain prosodic context while failing in other structures.

In order to perform a quantitative analysis of the patterns in reduced vowel

production across three populations, it is necessary to amass a sizeable corpus of recorded

utterances containing the target phone [ə]. This research design requires a large collection

of data samples from a demographically varied group of participants in order to minimize

the influence of any particular dialect or speaking pattern over the general results. The

research design further emphasizes the need to test participants in a variety of

phonologically-conditioned environments. The incorporation of multiple environments

ensures that emergent patterns are actually the result of variation in the production of

reduced vowels and not of any one prosodic feature. The purpose of this wide-scope

investigation is twofold: to predict spectral and temporal patterns of reduced vowel

production among monolingual, early bilingual, and late bilingual populations, and to

analyze the implications of these productions as they relate to studies of L2 perception.

29

2.1 Par ticipants

Statistical analyses of contemporary studies which were comparable in design

indicated that sixty-six participants, subdivided into equal groups of monolinguals, early

bilinguals, and late bilinguals, would be ideal for ensuring statistical significance of the

data. The decision was made to increase the number of participants to seventy-five in

order to further justify the results of the study as comparable to the general population.

For the purposes of this study, monolinguals were classified as native speakers of English

who did not acquire a second language prior to the “critical period” (Birdsong, 1999).

The acquisition of another language, such as learning French or Japanese in secondary

school, was not cause for disqualification for the study, provided the monolingual was

self-classified as living in a home where only L1 is spoken, received her formal education

in the L1, and that she speaks only L1 in the majority of social situations (Appendix A).

Early bilinguals are identified for the purposes of this study as speakers whose

native language is not American English and where L2 American English was acquired

before the age of ten. There is considerable disagreement regarding the boundaries of the

“critical period” (see Long, 1990; Birdsong, 1999). While Long maintains that L2

acquisition must begin before the age of six to acquire native-like pronunciation, other

researchers have set the boundaries at a more reasonable “before the onset of puberty”

criteria. For the purposes of the current study, the age of ten years was used as a cutoff

for early bilinguals in order to assure maximal distinction between early and late L2

speakers, while providing a comfortable cushion between the maximal age of these

participants and adolescence. Given the location of the experiment, metropolitan Miami,

it was found that the linguistic situation for the majority of early bilingual speakers

30

consisted of L1 Spanish acquisition in the home, followed by L2 American English being

acquired by the speakers sometime between pre-school and kindergarten. The mean age

for early bilinguals’ age of L2 acquisition was 4.76 years of age, which is consistent with

the criteria for classifying a speaker as a “pre-critical bilingual.”

Late bilingual participants are defined in the current study as those speakers who

began to seriously acquire American English at age fifteen or beyond. This

categorization provides a five year separation in age of acquisition (AoA) from the early

bilinguals, thus providing adequate separation between the early and late participant

groups. These participants were more difficult to collect due to the prerequisite that

participants read English sentences which were primed to change every three seconds.

Eliciting a conversational-rate of speaking through the rapid- reading of sentences

required L2 speakers who were both fluent and highly literate.

The majority of these twenty-five participants were of Cuban descent, with the

next highest groups being Venezuelan, Puerto Rican, and Spanish, respectively

(Appendix B, C). The average age of L2 acquisition for this group was 22.68 years. As a

consequence of the late average age of acquiring L2 American English, it follows that the

late bilingual group was significantly older as a whole than the early bilingual group,

many of whom were university students. The age difference is consistent, however, with

the range of monolingual participants, whose ages spanned from ages eighteen to sixty,

thus reconciling both bilingual test groups to the same demographic parameters set forth

by the control group.

31

2.2. M ater ials - I nstr ument I

It was determined that participants would read English sentences from a computer

screen in order to assess characteristics of specified unstressed syllables in the target

words. The first experiment measures spectral and temporal qualities of schwa across

three environments which are phonologically conditioned for schwa. In constructing the

instrument, target words were placed as either the first word in the sentence, or second

following only a single determiner or modifier (Appendix D). The placement was a

calculated effort to avoid the effects of haphazard word placement or the influence of

suprasegmental features. Sentences underwent interjudge reliability to ensure consistency

in the stress and intonation patterns of each construction. To elicit these utterances,

sentences were constructed which contained the following environments:

2.2.1. Plur al M or pheme

In this section the second word of each sentence was constructed with the

syllable-final phonological environment: “sibilant + plural morpheme {-əz}”, as in

“watches” [wɑɑəz] and “sashes” [sæɑəz]. All of the target words present the reduced

vowel in a post-tonic environment, where the schwa is nestled between two sibilant

consonants.

Ex: “Three judges took a vote.”

“These sashes tie in the front”

2.2.2. Possessive M or pheme

Target words containing the combination of “reduced vowel + possessive

morpheme” were further restricted to the construction “sibilant + possessive morpheme

32

{əz} as in the words “Marsha’s” [mɑɑɑəz] and “niece’s” [nisəz]. The use of proper

names created sentences in which it was sometimes necessary to place the target word

first in the sentence. With common nouns, such as “niece’s” and “judge’s,” the ordering

of “determiner/modifier + target word” was retained in order to assure maximum

similarity with the sentences containing plural morpheme targets.

Ex: “My niece’s team won the game.”

“Lisa’s twin looks just like her.”

2.2.3. R educed V owels in W or d-F inal Position

This section of the instrument tests words where the final syllable is unstressed

and is, therefore, mandatorily reduced as in “sofa” [sofə] and “ninja” [nɪ nʤə]. In this

section of the instrument words containing reduced vowels in word-final position were

placed in medial position within the sentence to avoid further lengthening of the target

sound. Furthermore, all target sentences were constructed to follow the target schwa with

coronal /t/ to control for variables in sound placement

Ex: “A trip to Russia takes money”

“The tuba takes strength to play.”

This section of the instrument was further divided into three phonetic categories

for the purposes of testing if the vowel qualities and the duration of word-final schwa are

affected by the preceding consonant. The consonants in question were divided into

distinct categories which represent three major places of articulation in American

English, namely: labial [f,b], coronal [ʤ, ʃ ], and dorsal [k,g]. While this sampling is

33

not inclusive of all the consonants in the aforementioned categories, these six phonemes

occur in markedly different areas of the oral cavity and can be seen as an adequate

sampling of the American English consonant inventory. Participants were presented with

two sentences each, comparatively similar in length and suprasegmental features,

containing the “target ‘point of articulation’ consonant + reduced vowel” environment.

Ex: “A ninja tiptoes in the dark.”

“Wear a toga to the party”

“A sofa takes two people to lift it.”

2.3. I nstr ument I I

The second experiment is a quantitative analysis of the production of word-

internal reduced vowels in environments characterized as “deletable” schwa or “non-

deletable” schwa (“DS” or “NS” here out). Deletable schwas are vowels which have

undergone multiple layers of reduction to become mid-central lax articulations due to

their position adjacent to a stressed syllable (“pretonic” or “post-tonic” throughout).

Following this process of reduction, the schwa may undergo optional deletion which

collapses the vowel nucleus, thereby reducing the number of syllables in the word by one.

Ex: [fӕmɪ li] [fӕməli] [fӕmli]

[prɑbobli] [prɑbəbli] [prɑbli]

In NS environments, the vowel may also be reduced to schwa in pretonic or post-tonic

position. However two criteria prevent the reduced vowel from undergoing deletion: the

vowel does not precede a second reduced vowel in the following syllable (as in

34

“testimony”), or the deletion of the reduced vowel would place primary and secondary

stress adjacent to one another in the word, which is in contradiction with patterns of stress

in American English.

Ex: [t stɪ móni] [t stəmóni] *[t stmóni]

[ɪ ʤɪnéʃɛ n] [ɪ ʤənéʃ ən] *[ɪ ʤnéʃ ən]

Semantically-related words were chosen from both DS and NS categories to measure

duration of the target schwa. Vowel quality measurements were also taken for the target

words in order to properly identify vowel substitutions where did not occur.

2.3.1. “ Deletable Schwa” I nstr ument

Six sentences were created using target words containing medial reduced vowels

that are phonologically conditioned to undergo optional deletion, as in “general”

[ʤɛnəɻ əl] / [ʤɛnɻə l]. All of the sentences were stated in the carrier sentence: “Don’t

say ____, say ______” to ensure that no prosodic features interfered with vowel

production.

Ex: “Don’t say principal, say first.”

“Don’t say probably, say maybe.”

Four control sentences were randomly interspersed throughout this section of the

instrument; however, they were not measured for any aspect of vowel production. These

control sentences did not follow the same intonation patterns as the target sentences.

After a pilot trial containing only the “Don’t say ____, say _____” sentences, it was

35

determined that control sentences would be necessary to vary the tempo of the readings

and to hold the attention of the participants.

2.3.2. “ Non-Deletable Schwa” I nstr ument

In accordance with the same pattern as the DS instrument, the words chosen for

this section contain reduced vowels that do not lend themselves to deletion. These non-

deletable reduced vowels occur in words which contain both primary and secondary

stress. Though a weak vowel, the schwas in these target words must maintain the

syllable nucleus to prevent adjacent primary and secondary stress syllables – the

positioning of such syllables is unfavorable in American English. The NS words chosen

for this section of the experiment are also semantically related to the DS category to

assure maximum phonetic similarity between the target words. Examples included

“generality” [ʤɛnəɻ æləɾ i] and “probability” [prɑbəbɑləɑi], as compared with

“general” and “probably” outlined above. These sentences follow the same pattern

“Don’t say ______, say ______” to free the target words of the effects of unwanted

interference (Appendix D).

Ex: “Don’t say principality, say town.”

“Don’t say testimony, say account.”

In keeping with the DS portion of the instrument, this NS section also contains four

control sentences interspersed throughout the instrument for the purpose of creating

variety in rhythm and preventing monotonous pronunciations by the speaker.

36

2.4 Pr ocedur e

Participants were situated alone in a quiet conference room with a personal

computer to read the instrument. The sentences were automatically timed to move

forward at three second intervals, where speakers were not required to actively progress

the instrument. It was the case that several participants were unable to complete entire

sentences at this speed; however, the data collection process was not jeopardized by the

speed of the instrument because of the early placement of target words in the sentences.

The rapid pacing of this instrument was deemed necessary to maintain the “normal

speech” tempo which is required for conditioned schwa deletion.

Utterances were recorded on a Sony ICD digital recorder and converted to MP3

files, which were then analyzed for temporal and spectral qualities using PRAAT speech

analyzing software. The F1 and F2 values for each vowel were determined by taking the

mean value of the center of each target nucleus. Measurements were interjudged for

reliabity with an agreed upon accuracy rating of 94% given the parameters of not more

than 50 Hz difference between F1 and F2 measurements, or not more than .05 ms

difference in duration. These measurements also contained an intrajudge reliability of

95.5%.

In analyzing the data, the average F1 (vowel height) values of the sibilant + plural

morpheme, sibilant + possessive morpheme, and word-final schwa environments were

ranked within each participant group to determine within-group tendencies. The same

step was repeated for the F2 values within each group, and the F2-F1 values (indicating

the relative position of a sound within the oral cavity) were also compared within each

37

group. Reduced vowels were isolated from the target word and measured for duration,

as well as F1 and F2 formants to identify vowel height and backness.

2.5 E xplanation of the pr ocedur e

The optimal environment for eliciting reductions in vowels is fast conversation

between people who share a relationship (Patterson, Locasto, and Connine 2003).

However there are severe limitations concerning a researcher’s ability to elicit specific,

semantically-related constructions. Coaxing bilinguals, particularly late bilinguals, to

produce these target words in the context of extemporaneous speech would not yield

consistent results. The present study acknowledges that obtaining speech productions

through the act of reading alters the rhythm and tempo of syllable production. Piske and

colleagues (2001) found that read speech has been judged as possessing a “stronger

foreign accent” when rated by monolingual speakers, perhaps due to the failure of many

bilinguals to receive formal education in L2 literacy. However, the productions elicited

by reading sentences fulfill the immediate purpose of providing comparative analyses of

the various qualities of reduced vowels within and between monolingual, early bilingual,

and late bilingual groups.

Presently, the accentedness of the production of reduced vowels elicited by

reading, as opposed to natural conversation, is less the focus of the current study than is

the variation which occurs between productions based on morphological differences.

This instrument allows for the comparison of all the aforementioned features by group,

and produces data which allows for a ranking of these features in a hierarchical manner

(with regard to duration, vowel height/backness). It is believed that the results obtained

38

from this technique accurately portray the similarities and differences in the way each

participant group perceives and produces reduced vowels in distinct environments.

2.6 M ethods of A nalysis

The first portion of the experiment, measuring duration and F1, F2 values of

schwa in the sibilant + plural morpheme {-əz} as in “sashes,” sibilant + possessive

morpheme {-əz} as in “Lisa’s,” and schwa in word-final position as in “Russia,”

compares results within-groups to determine which groups were able to differentiate

spectral and temporal characteristics in seemingly homophonous environments. The

statistical test chosen to determine the significance of vowel qualities/quantities within

groups was a one-way ANOVA repeated measures. An ANOVA test is designed to

evaluate the equality the means, with the ultimate aim of disproving the null hypothesis

(that the results between groups should show insignificant statistical variation).

Parameters for significance were set at .05, where means averaging less than this number

were deemed to be significantly different. F1, F2, and F2-F1 values (indicative of overall

positioning of the sound in the oral cavity) were assessed within-groups in this manner.

Regarding between-group analyses, a one way ANOVA test was applied to

compare the mean values of F1, F2, and F2-F1 in each phonological environment against

the same environment in the remaining populations. The purpose of this test is to

compare three or more independent groups where each group contains the same number

of participants and they are being tested on a variety of variables, all of which must

ultimately be judged independently. In certain cases, paired-sample T-tests were later

applied to two of the three populations for direct comparison of variables deemed to have

significant difference by the wide-scope ANOVA test. The purpose of including T-tests,

39

which produce redundant values, is to directly, and more simply, compare values of one

variable between two populations in order to make a clear and readable generalization of

the data. These t-tests alone would not be sufficient for data analysis, as the margin for

error greatly increases as multiple t-tests are run, as compared to the efficiency of a one-

way ANOVA repeated measures test, which also provides a usable effect size for each

variable in question.

The first portion of experiment also tested vowel duration using the plural,

possessive, and word-final environments. In keeping with the uniformity of statistical

testing, a one-way ANOVA repeated measures test was chosen to synthesize the results.

The “within-subjects effects” test also allows for data to be collected from the same

participants and subsequently measured across a variety of environments, with individual

differences between participants minimized because of the repetition of the

measurements. The ANOVA test was also used to compare the average duration between

populations. To determine the significance of variation in the mean lengths of duration,

Fisher’s Least Significant Difference (LSD) post hoc test was administered at .05 level,

denoting a 95% confidence interval.

While late bilinguals were originally excluded from cross-population comparison

in the duration portion of the experiment, due to perception and production errors which

posited peripheral vowels in lieu of schwas, the decision was later made to incorporate

the data through paired-sample T-tests directly comparing late bilinguals with early

bilinguals, and late bilinguals against monolinguals. It is important to recall that the data

presented in Chapter 4 regarding this section will likely compare unreduced vowels with

reductions performed by monolinguals and early bilinguals – however; the results of late

40

bilingual quantitative productions in any form shed light on the perception and

production capabilities of this particular population.

From the first portion of the experiment a further subsection of the data was

analyzed whereby word-final schwas were compared for both duration and vowel

qualities based on the places of articulation each reduced vowel follows. Post-labial,

post-coronal, and post-dorsal schwas were analyzed for length using a one-way ANOVA

repeated measures test on the three environments to determine whether variation in the

either spectral qualities or durations reached statistical significance. As a complement to

the ANOVA test, Fisher’s LSD post hoc test was also run with a standard confidence

level of .05, or 95% percent confidence interval.

The second experiment collected data from the bipartite categories of “deletable”

(DS) and “non-deletable” (NS) words 3

Ex: “testament” / “testimony”

“general” / “generality”

“probably” / “probability”

for the purpose of analyzing duration patterns

between groupings. Within this instrument, all of the target words consist of

semantically-related items in which one word allows for optional deletion of a syllable

and the other word does not.

The data in this section were analyzed using paired-sample t-tests to determine if

significant differences arise in the duration of these reduced vowels in each set of words.

3 For a complete description of the phonological environments conditioning “deletable” and “non-deletable” environments, see section 3.3.

41

The test is run only within-subjects for the purpose of determining which groups, if any,

place special emphasis on durational differences between the pairs. It is not necessary to

compare the durations between-groups; rather, it is the intention of this portion of the

experiment to determine whether early and late bilingual groups perceive and reproduce

the patterning established by monolinguals in differentiating between “deletable” schwa

and “nondeletable” schwa targets.

C H A PT E R 3. R esults and Discussion

Results of the experiment are presented below, given the data analysis procedures

outlined in section 2.6. Section 3.1 highlights the important findings regarding spectral

and temporal qualities of schwas occurring in the plural and possessive morphemes, as

well as in word-final position. These results are compared both as within-group analyses

and across groups to determine patterns of schwa pronunciation among all three

participant groups. Section 3.2 provides a comparison of the mean durations of deletable

and nondeletable schwas (DS and NS hereout) within groups and across groups to

determine if the difference reaches statistical significance. Section 3.2 further

deconstructs the data into semantically-related pairs in order to provide a clearer

explanation of the strategies employed by different speaker groups when pronouncing

medial schwa in a variety of morphological contexts. Section 3.3 provides useful

commentary regarding the collection and synthesis of data in the current experiment.

42

3.1 E xper iment 1

3.1.1. E xper iment 1(a) V owel Qualities C ompar ison

In comparing the duration of reduced vowels situated within plural morphemes,

possessive morphemes, and in word-final environments, a clear hierarchical pattern

emerges with regard to vowel height across morphological categories. Within each

population the F1 formant (height) is highest in word-final position, lower in the

possessive morpheme, and lowest when it occurs within the plural morpheme. Because

of the inverse relationship which exists between vowel height and F1 formants, the

results confirm that word-final schwas are lower vowels, plural schwas are produced

higher in the mouth, and possessive schwas fall somewhere in between. Figures 3.1.1a-c

provide an illustration of the variation in F1 vowel qualities production between the three

environments:

Table 3.1.a. Pairwise comparison of F1 values across monolinguals

*. The mean difference is significant at the .05 level.

Table 3.1.a displays a pairwise-comparison chart whereby each category (plural,

possessive, or word-final) is compared against exactly one other category in order to

determine if the differences in F1 values are statistically significant. The highlighted

43

regions demonstrate with confidence that there is statistically significant difference in the

vowel heights of each category in this population. For monolinguals, the estimated

marginal mean value for schwa in plural morphemes was 414 Hz, which is considerably

lower than canonical schwa values. Schwas in the plural morpheme average 431 Hz,

which is lower still than the mean of 490 Hz for schwa in word-final position. Values for

the monolingual control group may be summarized as follows (where the > sign indicates

a higher F1 value):

Word-final > Possessive Morpheme > Plural Morpheme

Table 3.1.b-c presents comparative analyses of the pair-wise contrast between F1 values

in all three environments for both the early and late bilingual groups.

Table 3.1.b. Pairwise comparison of F1 values across early bilinguals

*. The mean difference is significant at the .05 level.

The above comparison for early bilinguals indicates a pattern which is very similar to the

monolingual group. All F1 values between the plural, possessive, and word-final

morpheme are significantly different from one another. Interestingly, early bilinguals

display a mean F1 value of 414 Hz in the plural morpheme, which is identical to the

44

average value for monolinguals. The early bilingual participants displayed a remarkable

similarity to the vowel heights posted by monolinguals, as the possessive height of 436

Hz (compared to 431 in monolinguals) and 488 Hz in word-final position (490 in the

control group) show almost no variation in pronunciation. The situation diverges a bit

with regard to late bilinguals, as figure 3.1.c demonstrates:

Table 3.1.c. Pairwise comparison of F1 values across late bilinguals

* The mean difference is significant at the .05 level.

While all three categories easily reach statistical significance among the late bilingual

group, it is immediately apparent that the amount of variation in the F1 production of late

bilinguals is much higher than for the early bilinguals and monolinguals (p =.004

compared with p<.001). The standard error is also much higher, indicating that an

individual’s production of any given target schwa may vary significantly from the mean

value assigned to each section. The mean value for schwa in the plural morpheme was

480 Hz for late bilinguals, a value which is closer to canonical [ə] than the [ɨ ] sound

approximated by monolinguals and early bilinguals. Nevertheless, late bilinguals

continue to follow the hierarchical pattern for F1 values: word-final > possessive >

plural as outlined by monolingual speakers. While possessive schwa averages 504 Hz, the

45

word-final mean of 577 Hz indicates a blurred area of perception between [ə] and

Spanish [a], as some participants produced values in the 600+ Hz range to produce such a

mean. Figure 3.1.d plots the average F1 values for the plural morpheme, possessive

morpheme, and word-finally across all three language groups.

Figure 3.1.a Across-groups comparison of F1 values in plural, possessive, and word-final environments

In Figure 3.1.a, a fascinating pattern emerges with regard to target vowel height.

It appears that early bilinguals are able to very closely approximate the vowel height of

native speakers – a feat which is all the more impressive given the sociolinguistic context

of the experiment. Miami is a city populated by bilinguals across every generation.

Therefore, the logical assumption is that Miami natives, even those who learned English

at a young age, would have received bilingual, not monolingual native speaker,

phonological input in L2 learning. Yet, despite this less-than-ideal situation for L2

46

phonological acquisition, early bilinguals have perceived and produced reduced vowels

whose central height values are remarkably similar to those of native speakers.

The second vowel quality in question, described succinctly as the position of a

vowel on the “front/back” continuum in the oral cavity (identified by the phonological

feature [+back]), was measured by analyzing F2 formants. In spectral analysis, high F2

values coincide with front vowels, whereas low F2 vowels indicate that the vowel was

articulated farther back in the mouth. In vowel reduction, it is assumed that F2 will carry

a midrange value, which is consistent with the idea of vowel quality neutralization. In

the current study, F2 measurements in monolinguals carry an inverse relationship to F1

formants in that the highest F2 formants are found in plural morphemes, with

comparatively lower F2 formants in possessive morphemes, followed by the lowest F2

formants in word-final position as demonstrated by the hierarchy:

Plural Morpheme > Possessive Morpheme > Word-Final Position

This hierarchy was maintained by across all three populations, though several

interesting patterns were detected through statistical analysis. While monolinguals met

the expectation of statistically significant differentiation in the backness of vowels

between plural, possessive, and word-final environments, the F-value of this group was

much higher than in the other two populations. These high F-values indicate a wider

range of variability within the population with regard to vowel production. Specifically,

the F-value implies that the mean may not be indicative of the broad spectrum of F2

values produced by monolingual speakers.

47

Early bilinguals, on the other hand, showed surface differentiation between all

three phonological environments; however, when standard deviation is considered, the

significant difference between plural and possessive categories is nullified (p = .055).

Though nearly reaching significance, this p-value indicates that the production of reduced

vowels in plural and possessive morphemes is in reality quite close. The following

pattern is indicative of F2 hierarchical values in early bilinguals:

Plural Morpheme ≈ Possessive Morpheme > Word-Final Position

Late bilinguals pattern more like monolinguals in that there are significant differences

between the F2 values of schwa in the plural morpheme, possessive morpheme, and

word-finally. The major difference for this population is the rate of intragroup variation

for F2 values where F (2,48) = 29.659. This signifies a wide range of variability within

the late bilingual group. Essentially, this wide scope of variability reveals that, though

the means are comparable to monolingual values in each category, there is much less

uniformity regarding the relative backness of schwas in late bilinguals. Nevertheless, the

hierarchy of mean F2 values for late bilinguals is identical, through this particular

measure, to that of monolinguals:

Plural Morpheme > Possessive Morpheme > Word-Final Position

The final set of variables compared in Experiment 1(a) was the overall mean

measurement of F2-F1 formants, which plots vowel height against backness to identify

the quadrant of the oral cavity where a particular vowel production occurs. The location

in the mouth is then compared across groups to determine if the target vowels are being

48

produced in the same region of the mouth. Average F2-F1 values were compared

between populations using a one-way ANOVA repeated measures test to determine if

significant difference exists between vowels produced between the plural morpheme,

possessive morpheme, and word-final environments Table 3.1.a displays a cross-

linguistic comparison by phonological environment:

Table 3.1.d. Mean values of F2-F1 for monolingual, early bilingual, and late bilingual populations

On first inspection it is apparent in table 3.1.d that monolingual, early bilingual, and late

bilingual populations follow the same pattern with regard to hierarchical rankings of F2-

F1 values. In each participant group, the following trend is observed:

Plural Morpheme F2-F1 > Possessive Morpheme F2-F1 >

F2-F1 in Word-Final Position

It is noteworthy that monolinguals display slightly lower F2-F1 values in articulating

reduced vowels in both the plural and possessive morphemes, which fails to corroborate

Flemming’s (2007,2009) assessment that monolingual speakers articulate higher [ɨ ]

vowels in these positions. These numbers notwithstanding, it is possible, given the high

levels of variation (F-value), that many monolingual speakers do utilize this allophone

and that their productions have been overshadowed by the mean.

The pattern is a bit different however, with regard to F2-F1 values of schwa in

word-final position. Here monolinguals display a slightly higher mean F2-F1 (1250 Hz)

as compared with both bilingual groups (EB = 1163, LB = 1185.) These results likely

49

signify that monolinguals do not utilize as low of a vowel (F1) in word-final position as

do the bilingual groups. Perhaps the phonetic similarity between schwa and Spanish [a]

causes target undershoot, which is highlighted in this particular environment. The data

presented in 3.1.d, however, fail to support a significant distinction between linguistic

groups (p = 734, .772, .099 respectively). These numbers present a picture of relative

similarity in across-groups testing, which suggests that the most relevant factors

involving spectral quality variation occur within each individual group, and how each

group categorizes F1 and F2 values for a particular phonological environment is where

differentiation most likely occurs.

3.1.2. E xper iment 1(b) R educed vowel dur ation: A cr oss-gr oup compar ison

This section provides an across-group analysis of reduced vowel duration where

schwa occurs in either the plural morpheme, possessive morpheme, or word-finally. It is

the intention of this section of analysis to examine whether bilinguals who learned

English in childhood are able to make the same subtle alternations in vowel length that

monolinguals demonstrate, and to further see if the differences in production are

significant. Figures 3.1.e-g display a pairwise comparison of vowel duration within all

three groups to assess how each language group articulates vowel duration across

multiple environments.

50

Table 3.1.e. Pairwise Comparisons of average duration in monolingual speakers

*. The mean difference is significant at the .05 level.

The measure of difference in duration between the plural, possessive, and word-

final environments indicates that all three durations are significantly different from one

another in monolinguals. Schwa in the possessive morpheme carries the shortest average

duration (mean = .067 ms), followed by schwa in the plural morpheme (mean = .075 ms).

Predictably, the longest of the reduced vowels is word-final schwa (mean = .083 ms). A

Greenhouse-Geisser test further confirms significance where f (2,48) = 14.452 and p

<.001. Thus, the following pattern serves as a control for the bilingual test groups:

Word-Final Duration > Plural Duration > Possessive Duration

Early bilinguals display a remarkably similar durational hierarchy, with

possessive-morpheme schwas characterized as shortest in duration (mean = .068 ms),

followed by reduced vowels in the plural morpheme (mean = .076 ms). Word-final

schwa contains the longest duration (mean = .083 ms). The near-identical measurements

in duration are quite remarkable, yet it must be considered that aggregate data such as a

group mean does not provide a comprehensive portrayal of the variation in duration

51

spoken by early bilinguals and monolinguals due to dialectal differences, as well as

aforementioned social factors. These results give indication that early bilinguals are quite

sensitive to the patterns of reduced vowel duration produced by native speakers. Tests of

within-subjects effects are very similar to those of the monolinguals in that f (2,48) =

14.564 and p <.001.

Table 3.1.f Pairwise comparison of reduced vowel durations in early bilinguals

However, it is interesting to note that while the differences in duration reach statistical

significance between all three groups in the monolinguals, the differences between schwa

in the plural morpheme and word-final schwa not reach statistical significance (sig: .056)

though the actual means were very close in number (significance in this study occurs at

the .05 level). According to the analysis, temporal distinctions in early bilinguals should

be arranged as follows:

Word-final Duration ≈ Plural Duration > Possessive Duration

Late bilinguals present yet a third pattern of distribution for reduced vowel length

as evidenced by table 3.1.g. Within the late bilingual group, the difference in duration

52

between plural and possessive schwas is not statistically significant. Both plural and

possessive reduced vowels are statistically different from those produced word- finally.

Table 3.1.g. Pairwise comparison of the duration of reduced vowels in late bilinguals

*. The mean difference is significant at the .05 level.

Having failed to reach statistical significance between the duration of vowels in plural

and possessive environments, it is possible to encapsulate gradation between the three

environments as follows:

Word-Final Duration > Plural Duration ≈ Possessive Duration

A comparison of each category across groups, using a one-way ANOVA repeated

measures test, found no significant differences in the values of duration between

populations. It should be noted that this test produced relatively high standard deviations

(std dev. = .015-.030), suggesting that a lack of statistical significance does not imply that

there are no obvious differences in measurement between groups. What these tests more

concretely reveal is that the central issue in reduced vowel durations, for the designated

environments, emphasizes the three contrasting ways in which vowel durations are

53

ranked – a pattern which varied three times in as many populations. An inability to

perceive the slight alternations native speakers make in reduced vowel production may

well contribute to the failure of certain bilinguals to master larger prosodic features of

American English.

These alternations lead to a final question regarding vowel quality production in

these three environments: Do reduced vowels in word-final position behave differently

based on the phonological features of the consonant that they follow?

3.1.3 E ffects of pr eceding consonants on vowel dur ation in wor d-final position

The vowels in focus were also examined for vowel qualities and duration into the

following categories: post-labial vowels, post-coronal vowels, and post-dorsal vowels.

The aim of this experiment is to determine whether the CV# environment is affected by

the place of articulation which characterizes the syllable onset. Measurements were

taken from all three populations and compared within groups to determine if similar

duration patterns would be found in all three populations. Standard deviations for each

measurement range between .015 and .02, respectively. In the following figures 3.1.b-d,

post-labial (as in “tuba), post-coronal (as in “Russia”), and post-dorsal (as in “toga”)

vowel durations are presented for each participant group.

54

Figure 3.1.b. Mean duration for schwa in CV# position in monolinguals

While post-labial vowels are significantly different from post-coronal (sig. = .001) and

post-dorsal (sig. = .002), there is no statistical significance which differentiates post-

coronal from post-dorsal (sig. = .391, where the mean difference is significant at the .05

level).

As shown in figure 3.1.i, the same overall pattern displayed in figure 3.1.h holds

true for early bilinguals. Post-dorsal durations are longer than post-coronal, which in turn

are longer than schwas in post-labial position. Despite the apparent similarity to the

trends displayed in monolinguals, the standard deviations begin increasing in early

bilinguals, a trend which will be magnified in late bilinguals (std. dev. = .015-.026 for

each of the three places of articulation). This implies less uniformity in production,

though the overall pattern is quite similar to monolinguals.

55

Figure 3.1.c Mean duration for schwa in CV# position in early bilinguals

Though the increase in standard deviation tempers any predictions to be made regarding

word-final schwa duration in early bilinguals, it is interesting to note that early bilinguals

are the only participant group to reach statistical significance in comparing all three

phonological environments: post-labial, post-coronal, and post-dorsal. The least

significant difference in these environments for early bilinguals occurred between post-

coronal and post-dorsal environments (p = .026), though the difference is

overwhelmingly more significant than the same environments in monolinguals (p = .391).

The late bilingual data presented in figure 3.1.j patterns similarly to the monolinguals and

early bilinguals; however, as figure 3.1.k will illustrate, these are surface similarities

which do not carry the same implications for late bilingual vowel duration.

56

Figure 3.1.d Mean duration for schwa in CV# position in late bilinguals

While late bilinguals continue the pattern ‘post-labial < post-coronal < post-dorsal,’ the

durations are longer in each category than either the monolingual or early bilingual

groups. Because of the unusually high standard deviations in the measurements of late

bilinguals (std. dev. = .022-.044) none of the differences in duration between

phonological categories reaches statistical significance for late bilinguals. The high

standard deviations are indicative of widely variant patterns with regard to reduced vowel

duration, which implies that the pattern may not be as uniform as a simple comparison of

the means would suggest. Figure 3.1.k highlights the contrast in significance for late

bilinguals, so that a clearer picture may immerge.

57

Table 3.1.h Across-category comparison of word-final vowels in late bilinguals

*. The mean difference is significant at the .05 level.

The above chart attests, unquestionably, to the lack of statistical significance between

phonological environments when produced by late bilinguals. In the next section, the

results of vowel quality measurements of schwa in CV# position, following post-labial,

post-coronal, and post-dorsal consonants is synthesized.

3.1.4. E ffects of segmental featur es on the vowel qualities of r educed vowels in wor d-final position

The purpose of this portion of the experiment is to determine whether the spectral

qualities of word-final schwa move to assimilate to the location of the preceding

consonant. To make this determination, the F1 and F2-F1 values for each category were

compared within each linguistic group. After these measurements had been obtained, the

mean data from each C___# environment were compared across populations to provide a

comparative analysis of vowel quality production between groups. Figures 3.1.e-g

provide a comparative analysis of the mean F1 vowel qualities across all populations.

For a complete record of average values and standard deviations, see Appendix D.

58

Figure 3.1.e Across-groups comparison of F1 values in post-labial CV# environments4

In section 3.1.1 all participant groups were found to have the lowest F1 values in word-

final position. As figure 3.1.e displays, the mean value for F1 in post-labial ___ #

environments is 492 Hz, a relatively canonical value for schwa in monolingual American

English speakers. While late bilinguals had the highest F1 value in post-labial position,

this is to be expected due to the fact that late bilinguals had the highest F1 values in all

categories tested. Figures 3.1.f-g below present across-group comparisons of post-

coronal and post-dorsal environments.

4 Higher F1 values indicate that the target vowel is produced lower in the oral cavity.

59

Figure 3.1.f Across-groups comparison of F1 values in post-coronal CV# environments5

5 F1 values carry an inverse relationship to the position of the vowel in the oral

cavity. A high F1 value indicates the production of a low vowel and vice versa.

60

Figure 3.1.g Across-groups comparison of F1 values in post-dorsal CV# environments

The above figures present a clear picture of the comparative values of group’s mean

within an isolated environment. A further application of multivariate statistical tests

uncovers three contrastive patterns within the linguistic groups. In monolinguals, the F1

values, from least to greatest, may be interpreted as:

Post-dorsal > Post-labial > Post-coronal 6

In contrast, early bilinguals, who to this point have patterned monolinguals in many

aspects, display the pattern:

Post-dorsal > Post-coronal > Post-labial

6 “>” sign indicates a higher F1 value

61

Late bilinguals display an average pattern equal to that of monolinguals in the established

hierarchy of:

Post-dorsal > Post-labial > Post-coronal

A within-groups comparison, where all three places of articulation were compared

in one group only, returned several intriguing items. While monolinguals show no

significant differences between post-labial and post-dorsal articulations (p = .198).

Furthermore, the difference between post-labial and post-coronal environments is barely

significant (p = .048). Indeed, it is only the difference between post-coronal and post-

dorsal articulations that can be confidently stated to have predictable differentiation (p =

.000) When taken together, these numbers indicate that the difference in F1 vowel

qualities across these environments is not so significant in monolinguals as to rule out

coincidence. A similar phenomenon occurs in late bilinguals who show significance only

between post-dorsal and post-coronal articulations, and this difference was marginally

significant (p = .043). Conversely, all three places of articulation were significantly

different for early bilinguals. F1 values are useful points for spectral analysis because

they provide useful clues about which vowel[s] the speaker is targeting. In languages,

such as Spanish, where the phonemic inventory contains widely-spaced vowels, obtaining

an F1 vowel in isolation provides useful cues about the speaker’s perception of an L2

vowel. However, it is necessary to measure multiple spectral values for a vowel in order

to determine its overall location within the oral cavity. Therefore, it is obligatory to also

obtain F2-F1 values for schwa in post-labial, post-coronal, and post-dorsal environments

62

if a determination is to be made concerning the effects of target overshoot/undershoot on

the existence of a perceived foreign accent.

For the second section of this experiment, average F2-F1 values (indicative of the

overall positioning in the mouth based on the formula “backness-height”) were analyzed

to determine if the place of articulation [post-labial, post-coronal, or post-dorsal]

significantly affects vowel height or relative backness in the mouth. Figure 3.1.h

provides an across-groups comparison of these means for each participant group.7

Figure 3.1.h Descriptive Statistics of F2-F1 positioning across groups

At first glance, the F2-F1 structures display a more uniform pattern overall, as all three

populations tend toward a ‘post-labial < post-dorsal < post-coronal’ ranking of the F2-F1

7 For a complete list of the average means and standard deviations by place of

articulation, see Appendix F.

63

position. The noticeably higher F2-F1 vowels across post-coronal environments is

indicative of vowel fronting, a progressive assimilation technique caused by adjacency to

a coronal consonant. The application of a one-way ANOVA repeated measures test

further reveals several inconsistencies that the previous figure fails to elucidate.

With regard to the monolingual group, it becomes apparent that there is a failure

to reach statistical significance between the post-labial and post-dorsal groups (p = .129).

The same pattern is found among late bilinguals, who also fail to reach significance

between the post-labial and post-dorsal groups (p = .146). It should be noted that

significance between post-dorsal and post-coronal schwas, the late bilingual group barely

reaches significance (p = .048), which points to what is essentially a haphazard ordering

of vowel qualities among late bilinguals.

What is most intriguing among this data set is that early bilinguals, again, are the

only group to produce statistical significance between all three places of articulation. If

monolinguals were to consistently collapse two categories in terms of vowel quality

production, it is counterintuitive to believe that labial and dorsal areas of the oral cavity

would find themselves sharing coarticulatory features. Figure 3.1.i-k below illustrates the

relative position in the mouth of each group’s post-labial, post-coronal, and post-dorsal

means.

64

Figure 3.1.i Comparison of the position of schwa in CV# position among monolinguals.

As demonstrated in Figure 3.1.i the tendency of monolinguals is to produce a

significantly higher and more “fronted” schwa than occurs in either post-labial or post-

dorsal environments. While this is to be expected, it is interesting to note that the

difference between post-dorsal and post-coronal schwas did not reach statistical

significance, though both were significantly different from post-labial schwas. Figure

3.1.j compares these positions among the early bilingual population.

M post labial, 1185, 492

M post coronal, 1049, 487

M post dorsal, 991, 516

F1 H

z

F2-F1 Hz

M post labial

M post coronal

M post dorsal

65

Figure 3.1.j A comparison of the position in the mouth of schwa in CV# position among early bilinguals

Figure 3.1.j presents a remarkably similar picture to that of monolingual speakers.

These results establish with reasonable certainty that early bilinguals are capable of

perceiving mild differences in schwa production across a variety of post-consonantal

environments. It may be the case that early bilinguals overcompensate for these mild

differentiations, however, as all three CV# environments were significantly different only

in this group of participants. Figure 3.1.k illustrates the production values of these

schwas in CV# position as produced by late bilinguals.

EB post labial, 1185, 492

EB post coronal, 1049, 487

EB post dorsal, 991, 516

F1 H

z

F2-F1 Hz

EB post labial

EB post coronal

EB post dorsal

66

Figure 3.1.k A comparison of the position in the mouth of schwa in CV# position among late bilinguals

Upon first glance the pattern for late bilinguals appears to be the same. Truly, the

overall production pattern for late bilinguals mirrors that of monolingual and early

bilingual participants. However, in late bilinguals none of the categories, post-labial,

post-coronal, or post-dorsal, are significantly different from one another. This implies a

lack of perception with regard to the subtleties of variation in schwa production . These

three figures highlight two important generalizations regarding the phonetic environment

of schwa: The post-dorsal environment produces a lower vowel in all three participant

groups (as demonstrated by observably higher F1 values) and the post-coronal

environment invariably produces a more fronted vowel, as witnessed by higher F2-F1

values. Figure 3.1.l provides a conclusive visual representation of the variation one finds

in each of the average CV# environments across groups.

67

Figure 3.1.l Across-groups comparison of the average values of among three places of articulation

This figure clearly demonstrates that while the numbers across groups appear slight, an

overall pattern exists whereby monolinguals produce slightly higher and more fronted

schwas across all three environments, early bilinguals closely approximate these values,

and late bilinguals are produce schwas which are lower and farther back in the oral cavity

than either monolinguals or early bilinguals.

3.2. E xper iment I I

This experiment examines the duration of word-internal schwa in a variety of

intersentential environments. The target reduced vowels in this instrument (see Appendix

D) are classified as either “deletable” or “non-deletable.” Deletable schwas are vowels

which have undergone multiple layers of reduction to become mid-central lax

articulations due to their position adjacent to a stressed syllable. Reduced vowels are

particularly vulnerable to deletion when morphological alterations create adjacent

syllables containing a [ə] nucleus. The schwa may then undergo [optional] deletion

F1 H

z

F2-F1 Hz

Monolingual

Early bilingual

Late bilingual

68

which collapses the vowel nucleus, thereby reducing the number of syllables in the word

by one.

Ex: “opera” [ɑpəɻ ə] [ɑpɻə ]

“miserable” [mɪ zəɻ əbəl] [mɪ zɻə bəl]

By contrast, nondeletable schwas are vowels which reduce in pretonic and post-

tonic environments; however, the reduced vowel cannot adopt a null value. In many

environments, preservation of the syllable nucleus prevents adjacent primary and

secondary stress. Furthermore, nondeletable schwas do not occur adjacent to other

reduced vowels, and so their preservation contributes to the “galloping rhythm” of

American English.

The first section of this experiment analyzes patterns in the duration of deletable

and nondeletable schwas within and across all three linguistic groups. Results for the late

bilinguals are included, but consideration must be maintained of the fact that many, if not

most, of the target vowels produced by this group were closer to /a/ and /ʌ /. Therefore,

any intergroup comparison against monolingual or early bilingual /ə/ should be

interpreted with this in mind.

Section 3.2.1 provides an intragroup and intergroup comparison of the temporal

qualities of deletable and nondeletable categories of sounds. Upon removal of the control

sentences, included for the purpose of varying tempo and retaining attention to the

instrument, each group retained six target sentences which were uniformly phrased

“Don’t say ______, say _________.” (Appendix C) Utilization of a carrier sentence

controls for prosodic features such as stress and intonation which could influence the

69

duration of reduced vowel production. As is illustrated in section 3.2.2, this construction

also allows for an accurate comparison of semantically-related word pairs by removing

all extraneous features.

The second portion of this experiment seeks to determine whether the duration of

seemingly homophonous [ə] in semantically-related words (i.e., “probably” vs.

“probability”) differs significantly on an individual level. Results are analyzed by

splitting the target words off into semantically related pairs, whose schwa production is

analyzed for significance using paired-sample t-tests. It is believe that the uniform

conditioning of sentences within the instrument as “Don’t say ________, say ________”

allows for a legitimate comparison between word pairs due to the neutralization of

external contrasts between the environments.

3.2.1. C ompar ison of r educed vowel lengths in wor d-inter nal envir onments

This experiment is a within-subjects test to uncover whether measurably

significant differences in duration exist within each group. First the mean value for

monolingual duration of deletable and nondeletable schwas was calculated to determine

1) if a difference in duration exists between the two categories and 2) is this difference

statistically significant? Within this group, deletable schwas last an average of .035 ms,

compared with .045 ms for nondeletable schwas. The observable difference between

deletable and non-deletable schwas easily achieves statistical significance (p = .008).

Equivalent values were extracted from early and late bilingual populations, which are

displayed for comparison in figure 3.2.a.

70

Figure 3.2.a Across-group comparison of duration in deletable and nondeletable schwas

In each group, the difference between deletable and nondeletable schwas is

statistically significant. Early bilinguals average .040 ms in deletable environments,

compared with .047 ms in nondeletable, whereas late bilinguals display means of .041 ms

and .051 ms respectively. These means confirm a preference for shorter reduced vowels

in deletable situations. Furthermore, it appears from the data above that both deletable

and nondeletable schwas are uniformly longer in duration when produced by late

bilinguals.

The decision was made to compare the duration of deletable schwas and

nondeletable schwas between monolinguals and early bilinguals using a one-way

ANOVA test, performed in place of a paired-sample T-test due to the anticipated

similarity in results. Table 4.2.a illustrates the results of a comparison between the

monolingual and early bilingual groups.

71

Table 3.2.a. Comparison of deletable and nondeletable vowel durations across populations

ANOVA

This across-group comparison produces significant levels of difference in the “deletable”

group (p = .030), yet fails to reach statistical significance in duration for the “non-

deletable” category. This variation in reduced vowel production in deletable

environments may well be one of the indicators of a perceived foreign accent.

Late bilinguals were not compared for significant difference in duration against

monolinguals and early bilinguals due to the abundance of target vowel performance

errors noticed during the data collection process; the target substitutions in these cases

was often /a/, a vowel which, in comparison with [ə], would yield little insight into the

current research question. Nevertheless, it is worth noting that the average duration of

vowel productions for late bilinguals appears to be markedly different from that of

monolinguals, yet there is little discrepancy in the mean productions between early and

late bilinguals. Among all three participant groups the results indicate that [ə] is not

entirely homophonous in deletable and nondeletable environments with regard to

duration.

72

3.2.2. Semantically-r elated wor ds

In the final section of this experiment, word pairs ( semantically-related words

containing deletable/nondeletable items such as “general”/”generality”) were analyzed

using paired-sample T-tests. This test is applied to determine if the differences in data

from exactly two groups reaches statistical significance. This research question requires

a binary analysis of each pair to assess significant differentiations in duration of schwa.

Table 3.2.b presents the results of paired-sample t-tests8

Table 3.2.b Paired sample T-tests of semantically related word pairs

. This table highlights the

statistically significant findings between the aforementioned pairs. Interestingly, none of

the pairs shows statistical significance in the late bilingual group.

A lack of significance within the late bilingual group is likely the result of targetlessness

in the articulation of medial schwa. Furthermore, the complete lack of statistical

8 . For a comprehensive list of each word pair comparison, see Appendix E.

73

significance in any of the targets indicates a lack of methodical pronunciation of medial

schwa values. These indistinguishable values are completely divergent from the behavior

of monolinguals, who registered statistically-significant differences in five of the six

tested word pairs. Early bilinguals pattern somewhere in between monolinguals and late

bilinguals, with two of six tested environments (“testament”/ “testimony” and

“imaginative”/ “imagination”) reaching a level of significance.

3.3. Summation of the r esults

The six experiments in this chapter were conducted with the aim of determining which

patterns of bilingual speech varied significantly across populations. Patterns in variation

were confirmed most often in experiments which measured duration, rather than vowel

qualities per se. Statistical analyses indicate that the common factor among all of these

experiments is that the three participant groups pattern variables in different ways.

Whereas across-group comparisons may render differences too slight to register statistical

significance, differences within groups may provide insight as to how speakers will

synthesize target words within a given category. The following chapter provides final

commentary on these findings, which are incorporated into the larger discussion of L2

perceptual theory.

74

C H A PT E R 4. C onclusion

4.1. T heor etical I mplications for R educed V owel Pr oduction

Having amassed a considerable quantity of data concerning vowel reduction

processes, the current section seeks to integrate these data into two prominent theories of

L2 acquisition. Section 4.1.1 engages Flege’s Speech Learning Model (SLM hereout) in

a discussion of L2 achievement in bilinguals who have been L2 speakers for many years.

Section 4.1.2 examines the conclusions of this study within the framework of Kuhl’s

Native Language Magnet Model (NLM hereout). Kuhl’s model, outlining the

progression of L2 sound perception through perceptual mapping techniques, is analyzed

for compatibly with the current data, particularly that of late bilinguals (see Escudero,

2007).

Section 4.2 articulates several generalizations that can be made regarding

Spanish-English bilinguals’ production of reduced vowels. Section 4.2.1 details the

patterns one expects to find in vowel quality production across populations, whereas

section 4.2.2 addresses the overarching trends found in the current study regarding vowel

duration. Conclusions are drawn from the second experiment, highlighting medial schwa

in intersentential environments, in section 4.2.3. Final thoughts on the research design

and execution of the experiment, as well as recommendations for future study, conclude

the current study in section 4.3.

4.1.1. Speech L ear ning M odel

Flege’s SLM postulates that late bilinguals will experience greater difficulty

discerning the phonetic features of similar phones. In essence, rather than being

perceived as separate sounds, these non-distant segments may come to be processed as

75

“positional-define allophones,” and the distinction remains unlearned by the L2 speaker.

Flege does not propose a criteria for the boundaries of what constitutes a phonetically-

distant sound, thus creating ambiguity with regard to which sounds may qualify as

“similar.” While the lack of distance is obvious in sounds such as /t/ and / /, the

situation is a bit more complex with regard to reduced vowels.

The relationship between L2 target schwa and L1 vo

/ example. The complete neutralization of

vowel height and backness features creates an environment where multiple vowels are

acceptable candidates for L1 substitution. As a result of the absence of any vowel

occupying the central region in Spanish, it is likely that, as the SLM posits, L2 learners

do not perceive schwa as a central vowel. What is perhaps most interesting is that the

late bilingual speakers in this study did alternate the L1 vowel substitution to more

closely approximate the features displayed by native speakers. Specifically, in situations

where native speakers produced [ɨ ] (as in “judges”), late bilinguals were prone to

substitute the positional allophone [ɛ ], which is an allophone of /e/ that is produced

closer to the target [ə]. However, in word-final position late bilinguals who failed to

reduce the vowel uniformly substituted [a], which is in harmony with native speakers’

higher F1 values in this category.

Flege argues that perception problems are language-specific, and that it is the

available L1 phonemic options a speaker has to choose from which govern the L2

substitution. The current study supports this view in light of the coinciding of

substitution judgments and the Spanish vowel inventory made by the majority of late

bilinguals. As a final point to reconcile regarding the SLM, Piske, Mackay, and Flege

76

(2001) have found that L2 speakers who use their L1 infrequently pattern much more like

native speakers than L2 speakers who often employ their native language. The

hypothesis that people who use their L1 infrequently would perform better was not easily

testable in the current experiment because the location (Miami) requires most people to

use Spanish daily. This linguistic environment creates a situation where all of the late

bilingual participants engage in regular L1 usage – a factor which may have contributed

to lower success rates in L2 reduced vowel production.

4.1.2. Native L anguage M agnet M odel

In adapting her NLM to L2 acquisition, Kuhl proposes that the existence of an L1

perceptual filter creates difficulty because later learning is constrained by the initial

mappings ingrained in the neural structure (Escudero, 2007). As language learning

continues into adulthood, incoming L2 sounds will be drawn to the perceptual magnets

already ingrained in the learner’s brain. In order for new perceptual categories to form, a

sound must be distinguishable from all other preexisting magnets, otherwise perceiving

and acquiring the new segment are met with difficulty.

In brain imaging studies, the conclusion has been reached that only adult

bilinguals who acquire both languages early in life possess overlapping regions of the

brain when processing the two languages (Escudero, 2007). Therefore, Kuhl postulates

that adult bilinguals strive to maintain two separate perceptual systems to be activated

depending upon the speaker’s language mode. It is the opinion of the current study that

this failure to fully cement dual perceptual systems is less the cause of failure to reduce

vowels, than is simple L1 transfer of the [Spanish] premise that all vowels are full

77

vowels, and their articulation as such is integral to maintaining the staccato syllable

structure which is the preferred speech rhythm for late bilinguals.

4.2 G ener alizations r egar ding L 2 r educed vowel pr oduction

4.2.1. V owel quality distinctions

The current study supported several aspects of Flemming’s (2007, 2009) studies

on predictable distribution of certain vowel qualities in schwa. It was the finding of the

current study that all three participant groups distinguished between the plural morpheme

{-əz} being a higher central articulation (closer to [ɨ ]), with the possessive morpheme {-

əz} characterized by slightly lower F2 and higher F1 values. These predictable

differences support the rationale that “Rosa’s roses” are not truly homophonous words,

and that a distinction in transcription of these reduced vowels would more accurately

reflect these differences.

It is not the case, however, that the possessive morpheme {-əz} is simply “word-

final [ə] + /s/” as Flemming claims. Findings here indicate that word-final schwa was

characterized by higher F1 formants and lower F2 formants than one would find in

possessive endings such as the pair “Marsha”/”Marsha’s.” Higher F1 and lower F2

formants signify that word-final schwas are produced both lower and farther back in the

mouth than their possessive counterparts. The trend that word-final schwas are produced

lower and more centrally than the plural and possessive categories is generally consistent

in all three participant groups. Late bilinguals, however, have noticeably higher F1

formants (indicating a lower vowel) in this position which, coupled with longer duration,

argues for perception of word-final [a], particularly in cognates such as “sofa.” However,

late bilinguals follow the same pattern as native speakers in F2-F1 values: Word-final <

78

Possessive < Plural, leading to the conclusion that even adult learners perceive that schwa

in the possessive morpheme and in word-final position are separate constructions.

Monolinguals present a clear picture of the ideal relationship between F1 (height)

and F2 (backness) values in achieving native-like pronunciation. An inverse relationship

exists between the F1 values ranked “plural < possessive < word-final” and F2 values

which prefer “word-final <possessive <plural.” This inverse relationship is integral to

establishing accent-free pronunciation of the central vowels (and their allophones) in

American English. Figure 4.2.a provides a mapping of spectral measurements onto a

protypical vowel chart in SAE.

Figure 4.2.a Prototypical F1 and F2 values for Standard American English vowels

The chart above illustrates the fact that higher F2-F1 values [in plural and

possessive morphemes] in both bilingual groups indicates that these speakers may be

approximating mid-front vowel [ɛ ] instead of producing a truly neutralized value.

Having established this pattern of perception, the next step in the research was to

79

determine to what effect progressive assimilation would affect the vowel qualities of

schwa in word-final position. It became apparent from the onset that no participant group

recorded even F1 and F2-F1 values across the three groups: post-labial, post-coronal, and

post-dorsal. There were slight differentiations in the F1 values, as monolinguals and late

bilinguals followed the pattern: post-dorsal > post-labial > post-coronal, with early

bilinguals reporting post-dorsal > post-coronal > post-labial for F1. This section of the

experiment was exploratory, however, and to extract generalizations the instrument

would have to be greatly expanded.

Average F2-F1 values in word-final environments present a more unified pattern.

Schwas which occur in post-coronal environments display the highest F2-F1, which is

believed to be the result of vowel fronting due to progressive assimilation. Predictably,

post-dorsal environments displayed the lowest F2-F1 values, as velar consonants do not

facilitate fronting. This portion of the experiment was conceived as a subset of the larger

experiment on vowel qualities in plural, possessive, and word-final environments –

further research is necessary to establish conclusions on progressive assimilation in

schwa production.

4.2.2. V ar iation in schwa dur ation

The intricate variations each participant group employs with regard to vowel

duration became one of the most interesting parts of the study. Native speakers further

confirmed the hypothesis that plural {-əz} and possessive {-əz} are not homophonous by

reporting statistically significant differences as follows: word final > plural > possessive.

Early and late bilinguals do not distinguish these differences so clearly, nor do these

groups follow the same patterns amongst themselves. Early bilinguals do not distinguish

80

between word-final schwa and the plural morpheme, though possessive morphemes are

shorter. Late bilinguals, on the other hand, produce longer word-final schwas, but

significant difference between the plural and possessive morphemes is absent. The

inability of bilinguals to replicate native speakers’ patterns of difference in reduced vowel

duration may be a contributing factor to the perception of a foreign accent.

A few words are also in order concerning duration of the various subsets9

4.2.3. Semantically-r elated wor d pair s

of

word-final schwa. While all three groups conformed to the pattern post-dorsal > post-

coronal > post-labial, only early bilinguals had significant differences between all three

environments. Native speakers did not reliably produce a difference between post-dorsal

and post-coronal durations, though post-labial was invariably the shortest duration. It

may be the case that monolinguals’ propensity to truly reduce the features of word-final

vowels nullifies the distinction between dorsal and coronal, whereas speakers who are

only approximating vowel reduction are more vulnerable to the effects of surrounding

consonants. These assimilatory features are further examined in section 4.2.3, which

provides commentary on the durational aspects of medial schwa.

It was hypothesized at the onset of the study that monolinguals possess

subconscious phonological knowledge of deletion processes and would, therefore, make a

distinction in duration between an optional schwa ( deletable) and schwas which are

required to maintain a mandatory syllable nucleus (nondeletable). Since bilingual

9 Post-labial, post-coronal, and post-dorsal

81

participants did not delete schwas except in rare cases10

These test words were conceived of as semantically-related pairs in order to

control for all extraneous factors which could create sentential variation. The question

was then posed whether or not a significant difference arises in the production of schwa

in these phonetically-similar constructions. The results indicated quite divergent patterns

which have strong implications for the prediction of a foreign accent. Monolinguals

produced significant differences in duration in 5 out of 6 word pairs. Early bilinguals

produced significant differences in 2 out of 6 pairs, whereas late bilinguals produced

none. Alteration of this target [ə] provides the listener with a clue as to the ultimate

meaning of the word, thus aiding in intelligibility. Failure to provide this alteration

, it was unknown whether or not

they would be sensitive to temporal variations. As was expected, the study confirmed

that monolinguals indeed make significant distinctions between deletable and

nondeletable schwas. Surprisingly, however, this distinction was made within all three

participant groups. The question then became whether or not the production of these

values (i.e. monolinguals’ deletable duration vs. early bilinguals’ deletable duration)

differed across groups. In these tests, results attested to less uniformity, as only the

duration in deletable schwas which was significant was between monolingual and early

bilingual groups; none of the nondeletable differences were significant across groups. To

conclusively test whether or not there is a generalizable rule concerning deletable schwa

duration across groups, it is necessary to increase the size of the instrument and replicate

the experiment.

10 Early bilinguals often reduced “literal” to [lɪ trəl] and “general” to [ʤɛnrəl]

82

requires the listener to delay comprehension and results in the perception of a nonnative

speech pattern.

4.3. L imitations and E xtensions

The current study sought to examine reduced vowel production in Spanish-

English bilinguals across a variety of environments. As information was processed, other

interesting variables (such as the coarticulatory effects of the preceding consonant) came

into question, and were examined briefly in the course of this study utilizing participant

groups who were already assembled. This subset of the experiment should be considered

exploratory in nature, as it is imperative to amass more than two target sentences for each

environment in order to draw assess the validity of findings presented here. Further plans

for research include the expansion of the instrument of target sentences in

deletable/nondeletable environments from six sentences in each category in hopes of

defining a more refined phonological rule than “deletable schwas are shorter in duration

than those which cannot be deleted.” Future studies may wish to focus only on this

aspect of schwa production.

Another area of the study which may be altered in the future is the method of

eliciting these productions. Having participants read sentences from a computer screen

raises issues concerning their ability to produce sentences at normal speaking rates –

slower, more formal speech styles are not conducive to vowel reduction. Harrington

(2010) postulate that speakers make moment-by-moment decisions regarding the

listener’s need for information, and when this need is high (as in a formal experiment

setting) the speaker will increase articulatory effort. The use of the carrier sentence

“Don’t say _____, say ________” ensured uniformity in measuring spectral analysis;

83

however, this sentence also rendered target word prediction to be impossible, which these

authors postulate to result in hyperarticulation of the targets.

One method which has been recommended is the “delayed repetition technique”

whereby a participant listens to a prerecorded dialogue and then repeats one of the

recorded sentences when prompted. There is some concern regarding the possibility of

hindering late bilinguals by using this technique to assess reduced vowels, as the added

stress of having to answer correctly might cause additional stress to this participant

group. It will be interesting to note the outcomes of reduced vowel production under this

procedure.

Further study may also wish to focus on a demographic of bilinguals who are

more divorced from the L1 community. In Miami there is always the possibility that

bilinguals never received adequate L2 American English input to learn the sort of vowel

reductions which characterize monolingual speech. The pervasive bilingualism of this

city creates the possibility that L2 American English was acquired without proper input,

and that the situation was not remedied in a bilingual school and social environment.

Comparison of these results to those of bilinguals living immersed in the L2 should

provide more normative results. Nevertheless, the current study has been able to provide

several keen insights regarding the ability of high-functioning bilinguals to perform a key

phonological process of American English while in L2 mode. This study has also

proven, from a variety of measures, that early L2 acquisition is a predictive factor in the

ability of a bilingual to produce native-like segments and obey key phonological

processes as demanded by the ambient language.

84

References

Best, C. and Tyler, M. 2007. “Nonnative and second-language speech perception.” In O.S. Bohn and M. Munro, (eds) Language Experience in Second Language Speech Learning. Philadelphia: John Benjamins Publishing Company.

Birdsong, D. 1999. Second Language Acquisition and the Critical Period Hypothesis.

Mahwah, NJ: Lawrence Erblaum Associates. Browman, C. and Goldstein, L. 1990. “Targetless schwa: An articulatory analysis.”

In G. Docherty and D. Ladd, (eds) Gesture, Segment, Prosody. Cambridge University Press, 194-219.

Chomsky, N. and Halle, M. 1968. The Sound Pattern of English. New York: Harper

& Row. Dalby, J. 1986. Phonetic Structure of Fast Speech in American English. Bloomington,

IN: University of Indiana Press. Escudero, P. 1987. “Second language phonology: The role of perception.” In M.

Pennington, (ed) Phonology in Context. Palgrave Macmillan, 115-125. Flege, J. 1987. “The production of ‘new’ and ‘similar’ phones in a foreign language:

evidence for the effect of equivalence classification.” Journal of Phonetics, 15, 47-65.

Flege, J. 1991. “Age of learning affects the authenticity of voice onset time (VOT) in stop

consonants produced in a second language.” Journal of the Acoustical Society of America, 89. 395-411.

Flege, J. 1995a. “Second language speech learning theory, findings, and problems.” In

Strange, W., (ed) Speech Perception and Linguistic Experience: Issues in Cross-Language Research. Timonium, MD: York Press, 78-104.

Flege, J. 2005. “Origins and development of the Speech Learning Model.” 1st

ASAWorkshop on L2 Speech Learning. Simon Frasier University, Vancouver: April 14-15, 2005. Oral Presentation.

Flege, J., Bialystok, D., Mack, M., Sung, H., and Tsukada, K. 2006. “Degree of foreign

accent in English sentences produced by Korean children and adults.” Journal of Phonetics, 34, 153-175.

Flege, J. and Bohn, O.S. 1989. “An instrumental study of vowel reduction and stress

placement in Spanish-accented English.” Studies in Second Language Acquisition, 11, 35-62.

85

Flege, J., Frieda, A., and Nozawa, T. 1997. “Amount of native-language (L1) use affects the pronunciation of an L2.” Journal of Phonetics, 25, 169-186.

Flege, J. and Hillenbrand, J. 1984. “Limits on phonetic accuracy in foreign language

speech production.” Journal of the Acoustic Society of America, 76, 692-707. Flege, J. and Mackay, I. 2004. “Effects of the age of second-language (L2) learning on

the duration of L1 and L2 sentences: The role of suppression.” Applied Psycholinguistics, 25, 373-396.

Flemming, E. 2009. “The phonetics of schwa vowels.” In Minkova, D., (ed)

Phonological Weakness in English. Hampshire, England: Palgrave Macmillan Publishing, pgs in printing.

Flemming, E. and Johnson, S. 2007. “Rosa’s roses.” Journal of the International

Phonetic Association, 37, 83-96. Fokes, J. 1993. “The Elusive/Illusive Syllable.” Phonetica, Vol. 50, No. 2: 102-123. Hammond, R. 1999. The Sounds of Spanish: Analysis and Application. Purdue

University Press. Harrington, J. 2010. “Acoustic Phonetics.” In Hardcastle, W.J., Laver, J.L., and Gibbon,

F.E., (eds). (2010). The Handbook of Phonetic Sciences. 2nd edition. Wiley-Blackwell Publishing. 91-93.

Hillenbrand, J.M., Clark, M.J., Neary, T.M. 2001. “Effects of consonant environment on

vowel formant patterns.” Journal of the Acoustical Society of America, 109 (2), 748-763.

Ioup, G. 2008. “Exploring the role of age in the acquisition of a second language

phonology.” In Hansen-Edwards, J. and Zampini, M., (eds) Phonology and Second Language Acquisition. Philadelphia: Johns Benjamin Publishing Company, 176-203.

Kondo, Y. 1994. “Targetless schwa: Is that how we get the impression of stress

timing?” Proceedings of the Edinburgh Linguistics Conference, 1994, 63-76. Kuhl, P. 2000. “A new view of language acquisition.” Proceedings of the National

Academy of Science, 97, 11850-11857. Lenneburg, E. 1969. “On explaining language.” Science, 125, 635-643. Lindblom, B Björn. 1963. “Spectrographic study of vowel reduction.” Journal of the

Acoustical Society of America, 35 (11), 517-525.

86

Olive, J., Greenwood, A., and Coleman, J. 1992. Acoustics of American English: A Dynamic Approach. Springer Publishing.

Patkowski, M. (1990). “Age and accent in a second language: A reply to James Emil

Flege.” Applied Linguistics, 11 (1), 73-89. Patkowski, M. (1994). “The critical age hypothesis and interlanguage phonology”. In Yavas, M. (ed.), First and second language phonology. San Diego: Singular Publishing Group, 205-221. Piske, T., Mackay, I., and Flege, J. 2001. “Factors affecting degree of foreign accent in

an L2: a review.” Journal of Phonetics, 29, 191-215. Scovel, T. 1988. A Time to Speak: A Psycholinguistic Inquiry into the Critical Period

for Human Speech from Yavas, M. (2011). Applied English Phonology. Oxford: Wiley-Blackwell Publishing.

Stevens, K.N. and House, A.S. 1963. “Perturbation of vowel articulations by consonantal

context: An acoustical study.” Journal of Speech and Hearing Research, 6, 111-28.

Strange, W. and Shafer, V. 2008. “Speech perception in second language learners.”

In Hansen-Edwards, J. and Zampini, M., (eds) Phonology and Second Language Acquisition. Philadelphia: Johns Benjamin Publishing Company, 153-191.

Thompson, I. 1991. “Foreign accents revisited: The English pronunciation of Russian

immigrants.” Language Learning, 41, 177-204. Wright, R. 2003. “Factors of lexical competition in vowel articulation.” In Local, J.,

Ogden, R., and Temple, R. (eds) Papers in Laboratory Phonology VI: Phonetic Interpretation. Cambridge: Cambridge University Press, 75-87.

Yavas, M. 2011. Applied English Phonology. 2nd Edition. Wiley-Blackwell Publishing.

87

APPENDICES: Appendix A

Participants Gender Age Birth Place Age at Learning English

M1 Female 54 North Carolina native language M2 Male 50 New Jersey native language M3 Female 30 North Carolina native language M4 Male 60 Tennessee native language M5 Male 31 New Jersey native language M6 Male 33 Pennsylvania native language M7 Female 31 New York native language M8 Female 27 North Carolina native language M9 Male 33 North Carolina native language M10 Female 60 Maryland native language M11 Male 34 Maryland native language M12 Male 60 Missouri native language M13 Male 57 South Carolina native language M14 Male 29 Georgia native language M15 Female 50 North Carolina native language M16 Female 19 Florida native language M17 Female 19 New Jersey native language M18 Male 19 Florida native language M19 Male 19 Florida native language M20 Female 18 California native language M21 Female 23 Florida native language M22 Female 24 Florida native language M23 Female 24 Florida native language M24 Male 23 Florida native language M25 Male 18 California native language

88

Appendix B: Bilingual Demographics

E1 Male 25 Florida age 2 E2 Female 18 Florida age 2 E3 Male 23 Florida age 5 E4 Female 23 Florida age 3 E5 Female 22 Florida age 2 E6 Male 19 Florida age 5 E7 Male 18 Florida age 4 E8 Male 19 Venezuela age 6 E9 Male 20 Florida age 10 E10 Female 31 Georgia age 6 E11 Male 19 Florida age 5 E12 Male 20 Puerto Rico age 6 E13 Female 31 Chile age 9 E14 Female 18 Venezuela age 7 E15 Male 22 Venezuela age 2 E16 Male 18 Florida age 5 E17 Male 21 Florida age 7 E18 Female 18 Florida age 3 E19 Male 18 New York age 5 E20 Male 19 Florida age 2 E21 Male 18 Florida age 5 E22 Male 18 Florida age 5 E23 Male 20 Cuba age 5 E24 Male 26 Cuba age 2 E25 Male 30 Florida age 5

89

Appendix C: Late Bilingual Demographics

L1 Female 38 Argentina Age 24 L2 Female 48 Dominican Republic Age 16 L3 Female 36 Argentina Age 16 L4 Female 45 Cuba Age 15 L5 Male 22 Spain Age 16 L6 Male 26 Chile Age 18 L7 Female 37 Cuba Age 24 L8 Male 18 Cuba Age 15 L9 Male 50 Cuba Age 31 L10 Female 20 Spain Age 18 L11 Female 35 Spain Age 16 L12 Female 34 Cuba Age 15 L13 Female 24 Venezuela Age 23 L14 Male 25 Cuba Age 17 L15 Male 20 Puerto Rico Age 16 L16 Female 58 Puerto Rico Age 20 L17 Male 60 Puerto Rico Age 22 L18 Female 52 Cuba Age 36 L19 Male 19 Venezuela Age 15 L20 Female 24 Cuba Age 18 L21 Female 50 Cuba Age 25 L22 Female 35 Cuba Age 19 L23 Female 23 Venezuela Age 15 L24 Female 33 Peru Age 19 L25 Male 43 Cuba Age 24

90

Appendix D: Instrument

I. Experiment I:

A) Plural Morphemes:

1. My nieces talk for hours.

2. These crutches take some getting used to.

3. Three judges took a vote.

4. Their noses tickled from the smoke.

5. These sashes tie in the front.

6. Hot sausages taste the best.

B) Possessive Morpheme

1. Marsha’s talent is juggling.

2. Asia’s territory is vast.

3. The judge’s table was messy.

4. Lisa’s twin looks just like her.

5. Sasha’s teacher is German.

6. My niece’s team won the game.

C) Word-final Endings

1. A ninja tiptoes in the dark

2. A trip to Russia takes money.

3. You use a hookah to take in smoke.

4. Wear a toga to the party.

5. The tuba takes strength to play.

6. A sofa takes two people to lift it.

91

II. Experiment II:

A) Deletable Schwas

1. Don’t say opera, say concert.

2. Don’t say imaginative, say creative.

3. Don’t say principal, say first.

4. Don’t say testament, say document.

5. Don’t say general, say usual.

6. Don’t say probably, say maybe.

B) Non-deletable Schwas

1. Don’t say imagination, say creativity.

2. Don’t say principality, say town.

3. Don’t say operatic, say theatrical.

4. Don’t say probability, say chance.

5. Don’t say generality, say norm.

6. Don’t say testimony, say account.

92

Appendix E: Means and Standard Deviations for F1 values by place of articulation

Group Mean in Hz. Standard Deviation in

Hz.

Monolinguals

a) post-labial F1

b) post-dorsal F1

c) post-coronal F1

491.76

502.72

474.80

53.407

52.523

50.785

Early bilinguals

a) post-labial

b) post-dorsal

c) post-coronal

487.24

500.16

479.28

53.825

60.985

62.312

Late bilinguals

a) post-labial

b) post-dorsal

c) post-coronal

515.45

527.91

499.27

65.960

49.310

46.239

93

Appendix F: Means and standard deviations for F2-F1 values by place of articulation

Group Mean in Hz. Standard Deviation in Hz.

Monolinguals

Avg postlabial F2-F1

Avg postdorsal F2-F1

Avg postcoronal F2-F1

1185.44

1222.44

1348.84

179.100

142.730

101.211

Early bilinguals

Avg postlabial F2-F1

Avg postdorsal F2-F1

Avg postcoronal F2-F1

1049.16

1127.44

1299.80

199.364

172.537

168.715

Late bilinguals

Avg postlabial F2-F1

Avg postdorsal F2-F1

Avg postcoronal F2-F1

991.18

1090.64

1227.91

237.315

279.666

198.009